Академический Документы
Профессиональный Документы
Культура Документы
Abstract
This article addresses the subject of RF signal detection from the point of view of those who design
and calibrate impedance matching bridges and other measuring instruments operating in the HF and
VHF radio ranges. The principal discussion relates to the simple diode peak-detector; and covers
the different possible circuit configurations, the associated theory, and the numerical methods
needed for data analysis. Circuit techniques used to linearise the diode detector output are
discussed in a separate document (Part 2).
Analysis of the detector transfer characteristic for sinusoidal input shows that the AC-induced
error term involves the zero-order modified Bessel function of the first kind (I0). This result is not
new, but it is often disregarded. The dynamic contribution is quite unlike the error that occurs for
DC input, regardless of any compensatory modification of circuit parameters; which means that
linearity correction schemes using a reference diode to produce a DC amplifier with a
complementary gain law can never be perfect. It is also shown however, that the AC error is
independent of frequency provided that the smoothing capacitor is 'large'. This means that the
tracking detector system, which involves automatic self-calibration against a low-frequency
precision rectifier, is theoretically sound.
By considering the power dissipated in the detector, it is shown that the input impedance can be
calculated using first and zero-order modified Bessel functions of the first kind (I0 and I1). This
allows the determination of detector transfer-functions that take source impedance into account.
When this facility is combined with the ability to calculate the dynamic component of the peak
detection error, a measurement of DC output taken with a calibrated voltmeter can be converted into
a measurement of AC input without the need for an AC reference. The computation procedures
required are not simple, but they are described in detail and given as Basic algorithms readily
adaptable to any programming environment.
Table of Contents
Abstract.................................................................................................................................................1
Introduction..........................................................................................................................................4
1. Half-wave rectifier ..........................................................................................................................6
1.1 Circuit behaviour and basic design principles .........................................................................6
1.2 Diode peak inverse voltage ......................................................................................................8
1.3 Input chokes .............................................................................................................................8
1.4 Large signal input impedance ..................................................................................................9
1.5 Separation of port resistance and source impedance .............................................................11
1.6 Actual input voltage ...............................................................................................................12
1.7 Inductively loaded AM detector.............................................................................................13
2. Shunt-diode rectifier .....................................................................................................................14
2.1 Sampling floating voltages .....................................................................................................16
3. Voltage-doubler rectifier ...............................................................................................................18
3.1 Doubler input impedance .......................................................................................................19
4. Bridge rectifier ..............................................................................................................................20
5. Bi-phase rectifier ..........................................................................................................................24
6. Diode static voltage vs. current characteristic ..............................................................................26
6.1 Variation of saturable leakage current with temperature........................................................27
6.3 Diode measurements...............................................................................................................29
7. Signal diode data ..........................................................................................................................31
8. Diode circuit model ......................................................................................................................33
8.1 Diode stacking .......................................................................................................................33
9. Back diodes ..................................................................................................................................34
10. Vacuum thermionic diodes .........................................................................................................35
11. A brief history of diode detectors ................................................................................................36
12. Simple diode voltmeter dynamic characteristics ........................................................................41
12.1 AC-DC Transfer function .....................................................................................................41
12.2 Peak to average current ratio ...............................................................................................50
12.3 Detector power dissipation and input impedance ................................................................52
12.4 Diode power dissipation ......................................................................................................55
12.5 Fast unrestricted computation of input impedance ..............................................................58
12.6 Using diode dynamic resistance to estimate input impedance ............................................59
13. Calculation procedures for the simple diode voltmeter .............................................................61
13.1 Modified Bessel function, first kind, zero order ..................................................................61
13.2 Polynomial used in the asymptotic form, first kind, zero order ..........................................63
13.3 Determining output voltage from peak input voltage ..........................................................64
13.4 Determining peak input voltage from output voltage ..........................................................67
13.5 Derivative of the asymptotic form polynomial, first kind, zero order .................................70
13.6 Inverse modified Bessel function, first kind, zero order ......................................................71
13.7 Modified Bessel function, first kind, first order ..................................................................73
13.8 Polynomial used in the asymptotic form , first kind, first order ..........................................74
13.9 Ratio of modified Bessel functions, first order / zero order ................................................75
13.10 Determining output voltage from source off-load voltage ................................................76
13.11 Determining source off-load voltage from output voltage .................................................79
3
14. Generalised half-wave detector model .......................................................................................80
14.1 Series diode rectifier with port resistance and parasitics .....................................................80
14.2 Series-parallel to series transformation ................................................................................90
14.3 Shunt diode rectifier with port impedance and parasitics ....................................................92
15. Detector with diode series resistance .........................................................................................95
15.1 Instantaneous diode current .................................................................................................96
15.2 Average diode current by numerical integration ..................................................................98
15.3 Output voltage from peak input voltage ............................................................................105
15.4 Comparison of numerical integration and transformation methods ..................................107
15.x Peak input voltage from output voltage .............................................................................108
99. Work in progress ......................................................................................................................110
Acknowledgements
2 http://www.openoffice.org/
3 https://wiki.openoffice.org/wiki/Documentation/BASIC_Guide
4 http://web.mit.edu/soffice_v8.0/pdfdoc/staroffice-BASIC.pdf
Introduction
The diode detector finds widespread use as a high-frequency voltmeter; its principal advantage,
apart from the simplicity of the circuit, being the ability to provide a bandwidth of several hundred
MHz with minimal attention to physical layout. This property means that detectors can be
connected directly to the output ports of bridges and other measuring devices, thereby eliminating
the need for down conversion or high-speed sampling. The essential preconditions are that the
signal amplitude should be tailored to lie somewhere in the range from about 0.1 V to 10 V RMS,
and that the source network should have a moderate load-driving capability.
The difficulty with the diode detector however, is that there is a big difference between using it
to make a crude RF level indicator, and using it to make an accurate measuring instrument. This is
primarily because the circuit behaves in an extremely non-linear manner for small signals and
becomes only approximately linear as the signal voltage approaches the point at which the diode
will be destroyed. Correction is not straightforward, and widely-used linearisation schemes
involving a DC amplifier with a complementary gain law are not completely successful because the
behaviour under AC excitation is not the same as under static conditions. The matter of turning the
diode detector into an accurate broadband voltmeter is nevertheless well worth pursuing; not least
because the perfect solution, an active rectifier or 'superdiode' circuit, is typically restricted to an
upper frequency limit of about 50 kHz.
In this article, we will consider the diode detector on two ways: firstly, as a measuring device in
its own right; and secondly, as a circuit module for inclusion in more elaborate instruments. In the
latter case, we will assume that the implementation involves linearisation (or some other remapping) of the rectified output, using either digital or analogue methods. It is, of course, obvious
that any correction process is dependent on a detailed knowledge of detector behaviour, and so the
first part of the study is a necessary precursor to the second.
The archetypical diode detector is a half-wave rectifier, either driven by a transformer, or having
an RF choke as a DC short-circuit across its input. This can be analysed on the basis that he choke
or transformer is perfect, and results in the principal mathematical relationship that governs the
detector's dynamic behaviour. The solution (involving the modified Bessel function of the first kind
in zero order), although mentioned in the academic literature, is evidently unfamiliar to the majority
of commentators. It is therefore set out in detail in section 12.1. The insight it gives is central to
the matter of voltage measurement, but it is not sufficient to solve the overall problem.
The first issue is that the prototype detector to which the analysis relates is not necessarily the
best circuit to use. There are numerous rectifier configurations, with different properties and
idiosyncrasies, and the variants should be considered carefully before making a choice. This is the
material covered in sections 1 - 5. An important subtext to those considerations is that, in
maximising the bandwidth of the diode voltmeter, it is a good idea to eliminate inductive devices.
Doing that introduces resistance into the detector DC return path, and the resulting reduction in
efficiency must be taken into account.
In sections 6 - 11, we take a look at the subject of diodes. The main point is to make the reader
aware of the various types and their characteristics; although a considerable amount of background
material is included, for context, and to contradict historical misinformation. From all of that, the
best type of diode for a particular signal measurement applications should be clear.
A complete mathematical analysis of the prototype detector is given in section 12. This is set out
with all of the logical steps explained, so that it should be accessible to anyone with some
knowledge of calculus (although it is possible to skip to the result in each subsection). The
principal transfer function has already been mentioned; but in addition, it is necessary to derive
expressions for the detector input impedance. Only then does precision measurement become
generally possible, because knowledge of the input impedance allows the output reduction due to
source loading to be determined.
Once we have a complete knowledge of the detector transfer function, including loading effects,
5
the detector qualifies as an absolute AC voltage measuring instrument; at least in the limited sense
that an accurate DC measurement at the output can be converted into the peak value of a sinusoidal
input, all without recourse to an AC voltage reference. The next problem with which we are
confronted however, is that the computational procedures are not purely analytical; which means
that there is little to report in the way of handbook formulae. The solution is given in section 13, in
the form of a library of numerical routines. This turns out to be a sizeable collection, most of
which is involved with the calculation of modified Bessel functions and the circumvention of their
argument-range restrictions. For the purpose of modelling detectors however, or for converting raw
data into measurements, we can simply pass parameters to a main program, which uses the rest of
the library as a collection of service routines.
After all of that, we still only have an accurate knowledge of the behaviour of the simple
detector. Mercifully however, it transpires that we can take more realistic detectors, having source
DC resistance, parasitic reactances, diode leakage and series resistance; and aggregate the
parameters so that they can be passed to the simple detector programs. The necessary
transformations are described in section 14.
Although, in section 12, we have described the detector transfer function in terms of modified
Bessel functions, it is also possible to solve the integrals for average current and average power
numerically. This allows analytically intractable variants of the of the detector problem, such as
finite time constant (ripple), and diode parasitic resistance, to be included at the integration stage.
In particular, it transpires that while diode parallel resistance can be separated analytically, diode
series resistance ( Rds ) cannot. Therefore, in section 15, we set up the numerical integration method
for diodes with finite ohmic series resistance and compare it with the transformation method of
section 14.
1. Half-wave rectifier
The simplest and best known RF detector circuit is the half-wave rectifier. Despite its familiarity
however, the exact details of its operation are often poorly understood; not least because it is
frequently subject to the casual presumption that the DC resistance looking back into driving
network is irrelevant. It must be understood that the detector will not work if the source is a DC
open circuit (although there may be a weak response due to diode reverse leakage current); and the
AM demodulator on which the text-book explanation is usually based is driven by an IF
transformer, which is effectively a DC short-circuit. In voltage-sampling applications, there is often
a significant DC resistance, and this reduces the measured output. Therefore, in the circuit below,
the source network is depicted in a manner that gives an analytical separation between the DC
resistance (Rport ) and the output impedance (Zout ). Of course, in practice, the electrical components
involved will be common to both quantities, but separation for the purpose of predicting the circuit
behaviour is usually trivial. The symbol Rport is used incidentally, because it is often the DC
resistance of an RF sampling port on an item of test equipment, such as an impedance bridge.
1.1 Circuit behaviour and basic design principles
The source produces an on-load voltage Vin~ on its side of a hypothetical perfect coupling capacitor,
and this has no DC component. Vin~ is also the voltage that will be measured using an AC-coupled
probe at the input to an actual detector. It will be approximately sinusoidal provided that the input
resistance of the detector is large relative to the magnitude of the output impedance of the generator,
|Zout|. A DC offset is however added to the source voltage on the detector side of the hypothetical
capacitor, so that the actual input voltage is:
Vin = Vin~ - Vbk
where, presuming that the diode polarity is chosen to give a positive output; Vbk , the backoff
voltage, is negative relative to the ground rail shown. The backoff voltage arises because the
average diode current (Iav ) that gives rise to the DC output (measurement) voltage Vm must also
flow through Rport . Thus the DC signal produced by the rectification process is shared between the
output load resistance RD and the port resistance in proportion to the relative values of those
resistances. This means that the backoff voltage reduces Vm , for which reason Rport is shown
below the ground rail in the diagram. Vbk can be measured at the input terminal by placing an RF
choke in series with a high-input-resistance multimeter (provided that the choke has a high
impedance at the generator frequency). In some AM radio receivers also, an RF-decoupled
resistance is deliberately introduced in series with the final IF transformer output winding, so that
the backoff voltage can be used for automatic gain control (AGC).
To calculate the detector output for a sinusoidal input, we can define:
7
Vin~ = Vp Sin
where Vp is the peak voltage, and = 2f t is the time-varying phase angle. By this definition, Vin~
is the RMS of a sinusoidal voltage, so that:
Vp = Vin~ 2
The output of the detector is then:
Vm = Vp - Vf - Vbk
where Vf is the effective diode forward voltage drop under
dynamic conditions. For a semiconductor diode, Vf varies
approximately logarithmically with the average forward
current, but it also has a contribution given by the logarithm
of the the peak-to-average current ratio. The problem of
calculating Vf is investigated in detail in section 12; but as
a rule of thumb, it is worth remembering that in signal detection circuits involving average diode
currents in the region of 100 A to 1 mA it is about 0.2 V for germanium point-contact diodes,
about 0.35 V for silicon-metal Schottky diodes, and about 0.6 V for silicon P-N junction diodes.
Another way of considering the detector is to say that the total detected voltage is the sum of the
output and the backoff voltage. Thus:
Vm + Vbk = Vp - Vf = Vin~ 2 - Vf
When the port resistance is zero, the input voltage no longer has a DC component, and this
becomes:
Vm = Vp - Vf = Vin 2 - Vf
which is the more familiar expression for the output of a half-wave detector. Note that the output in
this case can approach the peak value of the input voltage, provided that Vin is large relative to Vf .
For this reason, rectifying detectors are often referred to as a peak detectors, even though the error
in the measurement is large in the absence of some form of linearity correction.
In the circuit diagram given earlier, the output load resistance RD is shown as being separate from
the meter, and the meter is assumed to be a perfect voltmeter, i.e., of infinite input resistance. In
practice, RD is often the finite input resistance of a voltage measuring or sampling device, or the
parallel combination of an input resistance and a load resistor. In a passive circuit, RD might be the
resistance of a voltmeter constructed by placing a resistor in series with a moving-coil
microammeter. For example, a 100 A meter padded-up to 100 k by means of a series resistor
makes a 10 V FSD (full-scale deflection) voltmeter.
The smoothing capacitor CD should be chosen so that the time-constant CD RD is at least 10
times the period (1/f ) of the lowest frequency at which measurements will be made. If the lowest
frequency is to be, say, 1 MHz and RD = 10 k, then we want CD RD to be greater than 10 s, i.e.,
CD >1 nF. If a meter is used as the indicator, it makes no practical difference if the capacitor is
somewhat larger than the minimum required, and so 0.1 F (ceramic disk) is a typical choice; but
very large capacitors will damp the meter response in a manner that depends on the driving network
8
output impedance. A slow response can be desirable if the signal is jittery, but in some applications,
such as finding the null point when adjusting a measuring bridge, a fast response is needed. 'Fast',
on a human timescale, implies a time constant of less than about 10 ms (i.e., < 1 F for a 10 k
load, < 0.1 F for a 100 k load, etc.).
Note incidentally, that if the signal generator is amplitude modulated (e.g., with a 1 kHz sine
wave), and the time-constant CD RD is chosen to be shorter than the period of the modulating
frequency (e.g., about 100 s or 10 k in parallel with 10 nF), then the indicator can be a highimpedance transducer (earphone, etc.), or an audio amplifier and loudspeaker. This arrangement
can be useful when the detector is attached to an RF bridge. Note that, if a magnetic transducer is
connected directly, a load equalising network is needed (see section 1.7)
1.2 Diode peak inverse voltage
The minimum safe reverse voltage rating for the detector diode (VRM ) is most easily determined by
starting with Rport = 0. In that case, for large inputs, CD is charged to a constant voltage approaching
Vin 2 , whereas the most negative instantaneous voltage appearing at the detector input is
approximately -Vin 2 . Hence the diode must have a VRM of at least 22 (i.e., 2.82) times the
maximum possible RMS input voltage.
Half-wave rectifier: Diode VRM > Vin (max) 22
Now if we allow Rport to be > 0, the output voltage Vm is reduced by an amount Vbk , but then a
quantity -Vbk is added to the most negative excursion of the input waveform. Hence the peak
inverse voltage to which the diode is subjected is not affected by the port resistance.
1.3 Input chokes
The loss of sensitivity that results from having a finite port resistance can sometimes be a drawback,
and the traditional solution to this problem (assuming that transformer coupling is not to be used) is
to place an RF choke across the input to the detector. The choke should have a reactance that is
high in comparison to the detector load resistance at the lowest frequency of operation. 3.5 mH is a
typical choice, and has a nominal reactance of about +40 k at 1.8 MHz. This however is a large
inductance in radio engineering terms, and such inductors have substantial self-capacitance. The
self-capacitance will resonate with the inductance at some frequency, and the reactance of the choke
will become capacitive at frequencies above that5. This leads some commentators to assume that
RF chokes should not be used above the self-resonance frequency (SRF), but this rule is misleading.
The SRF is actually the fundamental parallel resonance of the inductor, and corresponds to the
frequency at which the impedance becomes extremely large in magnitude. Hence, from an AC
point-of-view, the choke disappears at the SRF and so does not shunt the network. It also matters
little whether the off-resonance reactance of the choke is positive or negative, the necessary
criterion being only that it should have a large impedance magnitude in comparison to the other
impedances in the network. Things go horribly wrong however at the first series-resonant
frequency. This occurs at approximately twice the SRF, and corresponds to the point at which the
length of the wire in the choke is one electrical wavelength. At this frequency, and for a range
either side of it depending on the Q, the choke acts as a short-circuit and the output of the network
is seriously reduced.
The potentially idiosyncratic behaviour of the DC bypass choke is not necessarily an issue if the
inductor is well designed and the frequency-range is restricted, but spurious resonances cannot
5 RF Chokes, their performance above and below resonance, Courtney Hall, Ham Radio, June 1978, p40-42.
9
always be ruled-out without practical verification. In general, insofar as the use of chokes cannot be
avoided, an RF choke composed of multiple segments (pie-wound) gives the highest first seriesresonance and so is to be preferred. For a given length of winding wire also, a choke with a ferrite
core gives more inductance than a non-magnetically-cored coil, and so will have a higher set of
resonance frequencies. Many designers however, prefer to eliminate the choke wherever possible
(they are expensive, as well as troublesome), and accept the loss of sensitivity. If the source is a
complete DC open-circuit, the most basic solution is to place a resistor across the port; but this
shunts the RF signal and it is sometimes better to place resistance across other elements in the
network. A good design outcome is that which gives a port-resistance considerably smaller than the
detector load resistance.
1.4 Large signal input impedance
Both port resistance and diode forward voltage drop are sources of error requiring correction in
precision measuring applications; but there is a third cause of error that must also be considered.
This is the detector input impedance, which will obviously load the source network and cause its
output to droop.
At first glance, the problem of calculating the input impedance seems intractable; because the
diode does not obey Ohm's law and so the input resistance varies during the course of a cycle. For a
single-frequency component of the signal however (this being what we must consider when
determining impedance for the purpose of calculating the frequency response), it is sufficient to
have an average picture of the input impedance. This can be obtained by considering the power
delivered to the resistive elements in the network.
The AC excitation causes a direct current to circulate, and this causes the voltages Vm , Vf and
Vbk to develop. These are related to the peak voltage by definition (because Vf is actually defined
as the voltage error that occurs in peak detection). Thus:
Vp = Vin~ 2 = Vm + Vbk + Vf
where Vin~ is the AC component of the input
waveform considered in isolation from the
DC offset caused by Vbk . Also, in the likely
event that there is no physical separation
between the AC source network and Rport , it
is the voltage that would be measured using
an AC-coupled probe.
Now let us define the total resistance around the loop as Rtot , i.e.:
Rtot = RD + Rdiode + Rport
where, to a fairly good approximation (which we will examine in detail in section 12.3):
Rdiode Vf / Iav
Neglecting energy loss due to the circulating harmonic currents produced by the rectification
process; the total power delivered to the detector is:
Pin = ( Vm + Vbk + Vf )2 / Rtot
10
Thus:
Pin = ( Vin~ 2 )2 / Rtot
i.e.:
Pin = 2 Vin2 / Rtot
Now, if the input impedance is defined as RZin , we also have:
Pin = Vin2 / RZin
so that:
RZin = Rtot / 2
Thus, assuming that harmonic generation is a lesser aspect of the behaviour of the lightly-loaded
(voltage sampling) RF detector6, the input resistance for AC signals is half the total DC resistance.
This result might seem surprising, but it can be understood by thinking of the detector as a kind of
voltage transformer. The RMS average of a DC signal is the same as its ordinary average, and so in
the process of converting from AC to DC, the RMS level has been transformed-up by a factor of 2.
Thus the detector is a 1:2 voltage transformer, and hence a 1:(2) (i.e., 1:2) impedance
transformer.
It should be noted however that the detector input impedance is highly variable, depending on drive
level. This is made obvious by substituting for Rtot in the expression above:
RZin = ( RD + Rdiode + Rport ) / 2
The variable term is Rdiode , which is not a resistance in the normal sense, but a quantity related to
the average power dissipation in the diode (this is obtained by integrating the product of
instantaneous diode voltage and current over a cycle of the input waveform). If the detector input
voltage is large in comparison to Vf however, then we have:
RZin ( RD + Rport ) / 2
This is the limiting input impedance under large signal conditions (neglecting parasitic reactances);
and if Rport is zero due to the use of a choke or a coupling transformer, it reduces to:
RZin RD / 2
The limiting AC input resistance is a reasonable measure of the input impedance when the detector
is driven hard, for best linearity. It is therefore useful for gaining an idea of the worst-case source
loading. As the drive level is reduced however, due to the non-linear relationship between voltage
and current, Rdiode will increase and eventually come to dominate. Thus the input impedance
increases as the drive is reduced, becoming very large (practically an open-circuit) when Vin = 0.
In order to model the small-signal input impedance accurately, it is first necessary to develop a
6 This assumption does not cause significant discrepancies in data analysis when making measurements using lightlyloaded detectors. Also, the AC component of the effective diode forward voltage drop, as derived in section 12,
must account for the dissipation of most of the harmonic energy.
11
theory that describes the dynamic forward conduction characteristic. That matter is discussed from
section 12 onwards.
The limiting 1:2 impedance transformation rule applies to all non-voltage-multiplying detectors. In
the case of the bridge rectifier (section 4), the limit is harder to approach because the detector places
two diode forward voltage drops in series with the output. In the case of the bi-phase full-wave
rectifier (section 5) the limit is slightly easier to approach, because the average diode current is
shared between two diodes.
Note that an idiosyncrasy of half-wave rectification is that, since Rport is part of the source network,
the rectification process causes power to be dissipated in the source network. This is usually of
little consequence in signal-processing circuits; but it can be (or should be) an important
consideration in the design of power-supplies.
1.5 Separation of port resistance and source impedance
There is no hard and fast rule for the analytical separation of the generator output impedance and
the port resistance, save to say that is will usually be obvious. Take, for example, the circuit shown
below, which is Douma's bridge7, widely used for monitoring HF transmission lines and measuring
reflection coefficient.
In the circuit on the left, the generator is typically a radio transmitter, and the impedance Z is an
antenna system. The bridge combines a voltage sample, obtained from the capacitive potential
divider C1 , C2 , RV , and the output of a transformer in series with the line, which produces a voltage
proportional to the current. The circuit parameters are calculated so that the bridge gives no output
when the impedance Z is equal to the design load resistance of the transmitter. This balance
condition can be maintained over a relatively wide frequency range by correct choice of the
resistance Rv , which compensates for the falling reactance of the current-transformer secondary
winding at low frequencies. Rv is typically a few k. Thus, since the DC resistance of the current
transformer secondary winding will be very small by comparison, it should be obvious that Rport is
equal to Rv .
The output impedance at the generator frequency requires a little more consideration, but not
much. Generally, the bridge is designed so that it abstracts only a very small amount of energy
from the line. Thus the generator will hardly notice its presence, and its output will not droop
significantly as a result of the additional loading. We can therefore assume that the output
impedance of the generator is effectively zero. With that assumption; the output impedance of the
voltage sampling network (by Thvenin's theorem) is given by the parallel combination of its
resistance and reactances. The total output impedance is then obtained by placing the parallel
combination of the current transformer load resistance and secondary reactance in series with the
output impedance of the voltage sampling network. Thus:
7 Directional apparatus for use with high-frequency transmission lines. T Douma, US Pat. No. 2808566, 1957
(filed 1953).
12
Zout = ( RV // jXC1+C2 ) + ( Ri // jXLi )
This can be converted to R+jX form using the standard parallel-to-series transformations8.
1.6 Actual input voltage
Frequently, the RF design considerations for an item of test equipment lead to the relationship
between some quantity to be measured and the open circuit (off-load) voltage of an output port.
Since the port has a finite output impedance however, and the detector has a finite AC input
resistance; the detector will not see that voltage. Instead, the detector input voltage is the output of
a potential divider formed by Zout and RZin . Thus, if VS is the open-circuit source voltage:
Vin = VS RZin / ( Zout + RZin )
The voltages and impedances in this expression are,
of course, phasors; but since the detector does not
conserve phase information, all phasors that appear as
factors can be replaced by their magnitudes. Thus, if
we use the convention that a voltage not written in bold is a magnitude ( i.e.: V = |V| ):
Vin = VS RZin / | Zout + RZin |
Now, if we have computed Zout in the R+jX form, so that (say):
Zout = RS + jXS
then we can add the input resistance to it and work out the magnitude of the denominator in the
expression for Vin . Thus:
Vin = VS RZin / { ( RS+ RZin )2 + XS2 }
This equation is primarily a statement of the obvious, in that if we want to avoid error due to source
loading, it will be necessary to make RZin as large as possible and |Zout| as small as possible. Such
choices are not always available however; there being, for example, an issue when using a Douma
bridge to drive a moving-coil panel meter, because the bridge has a large capacitive reactance
component in its output impedance at low frequencies, and the meter will typically require 100 A
of drive for full-scale deflection (FSD). The only small consolation is that the source loading will
be reduced at low drive levels, because of the increase in the detector input resistance (as discussed
earlier). This unloading effect will partly offset the inherent non-linearity of the diode detector, it
might even be used as the basis for partial linearity compensation if the source impedance is mainly
resistive; but in general it will be frequency-dependent and hard to exploit to any advantage.
We have, in the preceding discussion, uncovered a number of potential sources of error affecting the
use of the simple half-wave rectifier for absolute RF voltage measurement. They are, so far,
quantifiable, and largely correctable; but they demonstrate that the behaviour of the detector is not
quite as simple as many circuit designers seem to assume.
8 See for example: AC electrical theory. D W Knight.. http://www.g3ynh.info/zdocs/AC_theory/ . Section 18.
Note that the parallel impedance ( // ) operator is defined so that: a // b = a b / (a + b) , where both a and b can be
real, imaginary, or complex.
13
1.7 Inductively loaded AM detector
Although the principal subject of this article is AC voltage measurement; introducing the
relationship between detector load and input impedance provides an opportunity to identify a
widespread design error that occurs when a detector used as an AM demodulator is operated with an
inductive load. This mistake typically occurs when a magnetic transducer, such as a telephone
earpiece or a pair of headphones, is driven by a detector, either directly or through a step-down
transformer. The issue is that the stated 'impedance' of the transducer is not the same as its DC
resistance. Typically, the nominal impedance will be the impedance magnitude at some mid-band
audio frequency, such as 1 kHz, whereas the DC resistance is likely to be smaller by a factor of
between about 1/5 and 1/20.
Thus, for example, if we were to connect a pair of 10 k high-impedance headphones to a
detector, it might turn out that the DC resistance of the 'phones is about 800 ; in which case the
large signal input impedance of the detector will be about 400 instead of 5 k. The effect,
assuming a finite source impedance, will be to introduce excessive audio distortion by preventing
the detector from moving out of its threshold region; and if the driving network is resonant, it will
radically reduce the selectivity for strong signals by reducing the circuit Q.
The defect is present in most traditional crystal-set designs9. It results in a peculiar behaviour
whereby there is relative silence between stations, but when a reasonably strong distant station is
tuned in, it is heard mixed with a strong local station.
The solution is to place a parallel RC network in series with the inductive load, as shown in the
diagram below. The network is known a a 'benny', after its inventor Ben Tongue10. Usual design
practice is to use the resistor RB to pad the transducer resistance to equal its nominal impedance
value, thereby establishing the detector large-signal RF input impedance. The coupling capacitor
CB is then chosen to have a reactance magnitude that is small relative to the transducer impedance
(say 1/10th) at the lowest required audio frequency.
Notice here incidentally, that the transducer T is shown with a step-down transformer, but a high-Z
transducer could just as well be used without a transformer. Also, the tuned input transformer is
shown with separate windings for the resonator and the detector, whereas they can be combined.
This sidesteps the question of whether or not it is a good idea to tap the detector takeoff point down
the tank coil to preserve selectivity. The answer is that it is not necessary to trade signal voltage for
selectivity if a benny is used, because the detector input impedance, and hence the loaded Q of the
filter, can be controlled by RB .
9 See, for example, The boy's book of crystal sets, W J May, Bernards Radio Manuals 1954.
http://www.rexresearch.com/xtlradio/boysbook.pdf (accessed 23rd Aug. 2014)
10 Crystal radio set design. Ben Tongue. http://bentongue.com/ (accessed 23rd Aug. 2014)
14
2. Shunt-diode rectifier
An alternative to the standard half-wave detector configuration is the shunt diode rectifier, shown
below. This is the prototype of the indirectly-grounded detector circuits favoured by the Collins
Radio Company in the 1950s, and popularised in SWR and impedance-measuring circuit
construction articles by Collins design engineer Warren Bruene11 and others12.
Note that a voltage placed in square brackets refers to the DC component of a waveform. As in the
previous section, Vin~ represents the AC component of the input voltage, whereas the actual input,
Vin , acquires a DC offset due to the rectifying action of the diode. As before also, we will define:
Vin~ = Vp Sin
where Vp is the peak voltage, and = 2f t , so that:
Vp = Vin~ 2
The circuit appears, at first glance, to place a diode across
the detector port; but it does no such thing because of the
action of the coupling capacitor CC . When Vin~ first goes
negative (relative to the polarities shown on the circuit
diagram), D1 conducts and clamps one end of CC nearly to
zero, except for a shortfall due to the diode forward voltage
Vf . This means that the most negative excursion of Vin is
clamped to -Vf . Then, since the charge in CC remains
substantially constant on the timescale of the AC signal, the
near-sinusoidal voltage at the cathode of D1 averages about
+Vp - Vf . The filter comprising the RF choke, Rport and CD
removes the AC component to give a measured voltage of
Vp - Vbk - Vf , which is the same as for the ordinary halfwave detector. Overall, the main theoretical difference between the ordinary detector and the shunt
diode detector is the order in which Vp , Vbk and Vf are added together.
Note that, as far as is possible, the nomenclature has been made consistent with that used for the
ordinary detector. Thus Rport , although it is no longer part of the source network, has exactly the
same effect on the DC output as does the DC resistance of the source network in the preceding
section.
For the filter that removes the AC component, we have the choice of using a choke, a resistance,
or a combination of the two (such as a realistic choke). As before, the use of a choke minimises Vbk
but introduces the possibility of parasitic resonances, and so if some reduction in sensitivity is
tolerable, the troublesome choke can be eliminated.
11 An Inside Picture of Directional Wattmeters, Warren B Bruene, QST, April 1959, p24-28
12 In-Line RF Power Metering, Doug DeMaw, QST, Dec. 1969, p11-16.
15
Input impedance
The AC input resistance of the shunt detector obeys the same 1:2 impedance transformation rule as
the simple detector, i.e.;
RZin = ( RD + Rdiode + Rport ) / 2
The filter components however are effectively in parallel with that, and the reactance of the
coupling capacitor is effectively in series; so that the total (neglecting parasitics) is:
Zin = jXCC + ( Rport + jXRFC ) // [ ( RD + Rdiode + Rport ) / 2 ]
If the coupling capacitor is very large, and the choke (as is often the case) is omitted, this reduces
to:
Zin = Rport // [ ( RD + Rdiode + Rport ) / 2 ]
The coupling capacitor can also be physically omitted if the driving network is open-circuit to DC.
If there is any DC path across the detector port however, CC is essential; and if needed it should be
chosen to have a reactance that is always small in comparison to RZin . If the lowest frequency of
operation is say 1 MHz, and RZin is about 100 k for large inputs, then the voltage drop across the
capacitor might be considered to be negligible if its reactance magnitude is (say) always less than
1% of RZin , i.e., less than 1 k at 1 MHz. Since C = 1/2f |Xc|, we can then calculate that CC
should be at least 159 pF. It makes no practical difference if the capacitor is somewhat larger than
the minimum required, and so 1 nF or greater would be a sensible choice. Note however, that the
detector response will become sluggish if the capacitor is very large and the return path has a
relatively high resistance, because it will take time for the circuit to reach equilibrium.
Diode PIV
Since the voltage at the cathode of the diode averages at roughly +{Vin} 2 (assuming that the diode
polarity is chosen to give a positive output), the instantaneous peak inverse voltage sustained by the
diode will be {Vin}22 . Hence, as in the case of the simple half-wave rectifier, the VRM for the
diode must be at least 22 times the maximum possible RMS input voltage.
Shunt diode rectifier: Diode VRM > Vin (max) 22
16
2.1 Sampling floating voltages
One of the great advantages of the shunt-diode detector is its suitability for sampling voltages
across network components that are floating with respect to ground. A practical example is given in
the diagram below, which depicts a circuit for determining complex impedance ( Z = R+jX ) using
only scalar voltage measurements. This technique was originally described by Doyle Strandlund13,
but the implementation shown below is due to M E Gruchalla14.
This circuit produces five DC voltages, which (assuming perfect linear rectification) are
proportional to the magnitudes of the AC voltages across the source, the reference resistor Rref, the
reference reactance Xref, the load Z, and the junction between the two reference elements and
ground (the output voltages are all referred to ground). The full set of voltages is such that, even
though the individual measurements contain no phase information, they can be combined in various
ways to determine any of the impedance or admittance-related attributes of the load, Z (including
the sign of the reactance or susceptance)15.
The diodes D1 , D3 and D5 all have their anodes connected to ground, and so the corresponding
detectors conform to the prototype circuit given earlier, except that they are intended to feed
operational amplifiers having high input resistance and so use resistive filters and dispense with the
chokes. Also notice that there is a 1 M resistor across each of the diodes, the reasons for its
presence in each case being firstly; that the detector must have a load if it is not to have an infinite
time constant (and connecting it across the diode avoids loading the output); and secondly; that the
op. amp. will require a few nA of input bias current, and it is a very bad idea to get that via the
diode. Note that the voltages produced by these three detectors will be positive with respect to
ground.
The detectors using the diodes D2 and D4 have the same design considerations as the others,
except that they measure AC voltages that float relative to ground. This is possible because the
13 Amateur measurement of R+jX, Doyle Strandlund, QST, June 1965, p24-27. [An ingeneous graphical method for
measuring complex impedance using only scalar measurements].
14 Complex Impedance Measurement Using only Scalar Voltage Measurements, M E Gruchalla, Communications
Quarterly, Oct. 1998, p33-43. [Analytical approach to Strandlund's method and other improvements ].
15 For a complete analysis, see: Impedance and admittance measurement using scalar voltage samples.
D W Knight. http://www.g3ynh.info/zdocs/bridges/
17
measuring circuit is isolated from the outside world by coupling capacitors, which means that the
AC signals are guaranteed to have no DC component. This allows the D2 detector to get its ground
reference via the 75 input shunt resistor, while D4 gets its reference via the 75 resistor and the
51 reference resistor in series. Since the detector currents are only of the order of a few A, the
anodes of D2 and D4 are effectively grounded from the DC point-of-view, and so the detectors give
positive outputs referenced to ground.
Another example of the use of shunt-diode detectors is given in the circuit below. This shows the
detector arrangement used in the MFJ269 antenna analyser16. A greatly simplified version of the
driving arrangement is also included to show that the circuit has complete DC isolation from the RF
source (and also, that it works at UHF).
16 http://www.mfjenterprises.com/
18
3. Voltage-doubler rectifier
The circuit shown below is obtained by the simple expedient of replacing the filter components of
the shunt-diode detector with a second diode. The effect, at least nominally, is to double the voltage
output in comparison to a half-wave rectifier, to eliminate the port resistance and hence the backoff
voltage, and to remove the need for an RF choke. The circuit can also be used for the detection of
floating AC voltages provided that the conditions outlined in section 2.1 are met.. Such are its
advantages that we might wonder why it is not universally adopted; but it turns out that its principal
strength is also its principal drawback in voltage-sampling applications. The issue is that its
effectiveness at converting AC input into DC output means that it has a surprisingly low input
impedance.
As before, we will define Vp = Vin~2 , where Vin~ is the AC component of the input voltage. The
operation of the circuit is then as follows:
When Vin first goes negative, D1 conducts and clamps one
end of the coupling capacitor CC to -Vf . This causes the
capacitor to be charged to Vp - Vf . The charge in CC then
remains substantially constant on the timescale of the AC
signal, and so the voltage across CC is placed in series with
Vin , ultimately causing the smoothing capacitor CD to be
charged, via D2 , to twice the peak AC input minus two
diode forward voltages. Note that since the diodes operate
alternately, this detector conducts on both positive and
negative half-cycles of Vin , which means that it is actually a
type of full-wave rectifier. Also notice that the detector has
an output shortfall of two diode forward voltages; but since it also give twice the output of a halfwave detector, the overall degree of non-linearity is (at least nominally) the same as for the halfwave detector .
Diode PIV
Due to the clamping action of D1 , the voltage at the anode of D2 cannot fall substantially below
zero. Hence, since the voltage across the smoothing capacitor CD is very nearly +Vin 22 , VRM for
D2 must be at least 22 times the maximum possible RMS value of Vin . In the case of D1 , since its
effect is to clamp the most negative excursion of the input voltage approximately to ground,
thereby charging CC to Vin 2 , the voltage across CC will be placed in series with Vin~ when it
makes its positive excursion. Hence VRM for D1 must also be at least 22 times the maximum
possible RMS value of Vin .
Voltage doubler: Both diodes, VRM > Vin (max) 22
19
3.1 Doubler input impedance
In order to determine the input impedance of the voltage doubler, we can use an argument similar to
that given in section 1.4.
Provided that the loading on the source is not so great as to cause substantial harmonic
distortion, most of the harmonic energy generated by the rectification process will be dissipated in
the dynamic component of the diode forward voltage (section 12). In that case, the DC power
dissipated in the network will be the same as the AC input power. The total DC voltage produced
is:
2Vp = 2 Vin~ 2
and the total DC resistance is:
Rtot = RD + 2Rdiode
Hence:
Pin = ( Vin~ 22 )2 / Rtot = 8 Vin~ 2 / Rtot
but, by definition:
Pin = Vin~2 / RZin
Hence:
RZin = Rtot / 8
and for large signals, when RD >> Rdiode ,
RZin RD / 8
Thus, whereas the half-wave detector is a 1:2 impedance transformer, the voltage doubler is a 1:8
impedance transformer.
The disadvantage of the voltage doubler17 for non-invasive voltage-measurement purposes can now
be demonstrated by a simple example. Suppose we decide to use a voltage doubler to drive a
100 A moving-coil meter padded to 100 k to give 10V FSD. Neglecting diode resistance and
parasitics, the detector input impedance will correspond to a resistance of about 12.5 k. If we
were to use a half-wave detector however, we would get half the output voltage, but we could easily
reduce the meter padding resistance to 50 k (including Rport) and thereby increase the sensitivity to
5 V FSD. This alternative half-wave detector will now give exactly the same meter readings as the
doubler, except that its large-signal input impedance will be about 25 k. We have not,
incidentally, found a source of free energy by using a non-multiplying detector. We have merely
avoided pumping the smoothing capacitor up to a high voltage, only to drop the voltage back down
again using a resistor.
17 See also: Crystal radio set design technical help, Ben Tongue,
http://bentongue.com/xtalset/0def_exp/0def_exp.html
20
4. Bridge rectifier
A non-multiplying detector that requires no RF chokes, does not have a backoff voltage, and does
not care whether or not there is a DC path through the source network, is the full-wave bridge,
shown below:
This circuit is nearly universal in modern power supplies, but it has two disadvantages in signal
detection applications; one being that it has two diode forward-voltages in the path to the load, and
the other being that either the output or the input terminals must be allowed to float with respect to
the system ground.
Despite the caveats, one advantage of the bridge rectifier is that the maximum inverse voltage
for any of the diodes is only 2 times the RMS input voltage. The magnitude of the inverse voltage
across D2 is is prevented from rising above Vin 2 by the clamping action of D1 and vice versa. The
same argument applies for D3 and D4 .
Bridge rectifier: All diodes, VRM > Vin(max) 2
Principle of operation
Although the bridge rectifier is very familiar, its mode of operation is surprisingly difficult to
understand. This can be made apparent by asking the question: 'If the circuit above is grounded as
shown, what voltage waveforms, measured relative to ground, will appear at points V1 and V2 ?' If
you have not previously needed to solve this problem, it will probably take you a while to work out
the answer.
A simple way to find the solution is to imagine adding centre taps to both the input network and
the load network, as shown in the diagram below.
21
It is not really necessary to tap into the smoothing capacitor as well as the load resistor, but if we do
so the circuit becomes reminiscent of the familiar dual-rail power supply18. All we have to do to
turn it into the dual-rail supply is move the earth to the centre tap and feed it with a centre-tapped
transformer instead of a resistive potential divider. Thus it should be obvious that if the two halfoutputs are equally loaded (which they are), there will be no current flowing along the connection
between the input and output centre taps. This means that, with the negative terminal of the output
grounded, the alternating voltages V1 and V2 must be symmetric about a DC offset at half the
output voltage. So now, if we let:
Vin = Vp Sin
Then:
V1 = ( Vm + Vp Sin ) / 2
and
V2 = ( Vm - Vp Sin ) / 2
Note that these two voltages are identical apart from a 180
phase difference. Also, the AC component in each case is half
the input voltage.
Now compare the graph of the operation cycle given above with the corresponding graph for the
voltage doubler in section 3. It appears that the bridge circuit splits the input voltage into two
halves in antiphase. These two half-voltages are then applied simultaneously to a pair of voltage
doublers. Since there is no overall voltage multiplication however, the input impedance is:
RZin = ( RD + 2Rdiode ) / 2
Ripple frequency
The rectifiers described in sections 1 - 3 all top-up the smoothing capacitor once per cycle. The
bridge rectifier however, tops up the capacitor twice per cycle, which means that it doubles the
ripple frequency. Also, if the circuit is used without smoothing, it becomes a highly efficient
frequency doubler, removing the fundamental frequency-component completely (assuming perfect
symmetry). The frequency-doubled output is not sinusoidal, of course, but it can be filtered.
The frequency doubling effect is useful in mains-frequency power-supply applications because it
reduces the required smoothing capacitance for a given level of ripple. It is however, usually of
little importance in radio-frequency applications because smoothing capacitors are, in any case,
small.
The bi-phase (half-bridge) rectifier circuit discussed in section 5 also doubles the ripple
frequency; but note that the voltage doubler (section 3) does not, even though it is a type of fullwave rectifier.
18 It turns into a pair of bi-phase rectifier power supplies of opposite polarity. See section 5.
22
Frequency sampling
The fact that the input terminals of a bridge rectifier, when measured relative to ground, carry an
AC signal of half the floating input voltage can be exploited for the purpose of sampling the input
frequency. This is the traditional method for extracting the power-line frequency from mains
transformers without using a separate winding, and is used to provide a simple and accurate timing
reference in devices such as electronic clocks.
A power-line sampling circuit is shown on the
right. The forward voltage for silicon P-N
power diodes is about 0.7 V, so the powersupply produces a raw DC output of
7.52 - 20.7 = 9.2 V, which is intended for
regulation to 5 V for powering logic circuitry.
Thus the voltage to ground on either side of the
AC input to the rectifier is 3.75 V RMS, with a
DC offset of +4.6 V. This corresponds to a
peak-to-peak excursion of 3.75 22 = 10.6 V,
but it is fed through a resistor and a coupling
capacitor to a peak clipper comprising two
1N4148 silicon P-N signal diodes. These have
a forward voltage of about 0.6 V, and so the
resulting waveform is limited to lie between
0.6 V above the +5 V rail and 0.6 V below
ground, i.e., it is clipped to 6.2 V P-P. This is
converted into a clean 5 V P-P square wave by
means of a CMOS Schmidt trigger.
It should be understood, of course, that any substantial loading of the voltage at the AC input
side of a bridge will upset the symmetry. For a circuit designed to produce hundreds of milli-amps,
or even amps, of output current however, a sampling network such as this will make no measurable
difference. Note that the 100 nF capacitor has a reactance of -32 k at 50 Hz (-27 k at 60 Hz),
and although there will be harmonics present, it constitutes a very weak coupling. The 1 k resistor
gives the diodes and the Schmidt trigger some protection from any fast transients that might appear
on the power-line.
23
Coupling capacitor
Because the bridge rectifier introduces no net DC component into the current flowing in the source
network, it can be fed via a coupling capacitor (or capacitors) if so desired. In power supplies
however, the use of a coupling capacitor can be dangerous unless the implications of so doing are
considered in advance.
In the thoroughly nasty circuit shown on
the right, a bridge is fed via a coupling
capacitor, and four switches are provided for
the purpose of simulating a short-circuit
(which is the usual failure mode) in any one
of the four diodes.
Let us first consider what happens when
diode D3 is shorted. In that case, V2 is pulled
to ground, D1 and D2 begin to act as a voltage
doubler, and D4 is harmlessly reverse-biased
(it is assumed, incidentally, that none of the
diodes have their VRM ratings exceeded when the event occurs). From an external point of view, the
effect is approximately to divide the input impedance by a factor of 4 (assuming a purely resistive
load), and to double the output voltage. Shorting-out D1 will, of course, have a similar effect.
Now consider the effect of shorting-out D2. In that case, V1 is tied to the positive output
terminal, D3 and D4 act as a voltage doubler producing an output that is negative relative to the
positive terminal, and D1 is permanently reverse biased. Shorting out D4 will have a similar effect.
Thus, when the bridge is capacitively coupled, there are four ways in which the most common
type of diode failure can double the output voltage and quarter the input impedance. A fuse is
generally too crude a safety device to detect such an impedance change, and circuitry subjected to
twice its intended operating voltage can malfunction in numerous unpleasant ways. Such might be
the consequence of failing to consider this idiosyncrasy. On the other hand, a voltage doubler
circuit can be provided with a half-output () facility by the addition of two more diodes an a
single-pole switch.
24
5. Bi-phase rectifier
The bi-phase (or 'half-bridge') rectifier was once widely used in the power supplies of valve (tube)
equipment. As a power rectifier, it is a poor choice in comparison to a full bridge; and after
selenium HV rectifiers19 became available in 1933, its chief design merit seems to have been that it
maximised the profits for owners of light-bulb factories while increasing the size and complexity of
the mains transformer at the customer's expense20 21. The circuit creates a serious fire hazard in the
event that one of the rectifiers should fail open circuit22, because a DC component will then be
present in the transformer secondary current, causing core saturation and overheating.
In signal detection applications however, the circuit has the advantage of being a full-wave
rectifier with only a single diode forward-voltage drop in the path to the load. By conducting on
both half-cycles of Vin , the circuit also halves the average diode current for a given output voltage,
and so reduces the diode forward voltage drop compared to that of a half-wave rectifier. Because of
the logarithmic diode V-I characteristic however, the Vf reduction is only a few %.
A disadvantage of the circuit that it usually requires a centre-tapped coupling transformer (although
it can be constructed as a pair of half-wave rectifiers driven by a dual-ended (balanced output)
amplifier). The output voltage of the transformer-coupled version is given approximately by the
following expression (which neglects transformer losses and diode forward voltage drop):
Vm = Vin (2) Ns / Np - Vf
Where Ns / Np is the turns ratio. Correct operation requires that the transformer has sufficient
primary inductance not to load the driving network significantly at the lowest frequency of
operation, and that the transformer is otherwise working within its pass-band. For HF radio
frequency applications, the transformer can be wound on a small ferrite toroid or two-hole
('binocular') core. Note incidentally, that if a transformer is used for the ratio arms (i.e., the voltagedividing network) of a measuring bridge, it is possible to combine this transformer with the biphase detector transformer by using the fact that the bridge is a linear reciprocal network (i.e.,
generator and detector ports are interchangeable). The resulting configuration, a type of
19 Selenium high-voltage rectifiers (see: http://en.wikipedia.org/wiki/Selenium_rectifier) were considerably more
efficient than valves and require no heater supply, but equipment manufacturers associated with the valve cartels did
not use them. The slow adoption of semiconductors in some quarters is further discussed in section 11.
20 As pointed out by Bob Weaver (private e-mail, 14th Aug 2014), the weight of copper required is not greatly
increased when a bi-phase is used instead of a bridge, because the two secondary windings provide half the current
each. The manufacturing time and the amount of insulating material however, will be increased, and the valve
rectifier (in safe design practice at least) requires an additional floating heater winding.
21 The marketing of radio receivers in the 1930s, particularly in the USA, led to tube wars. Using more tubes was
supposed to make a better radio set. Thermionic rectifiers were counted as tubes, even though they were nothing to
do with the signal chain, but the semiconductor rectifier was not so easily turned into a marketable concept.
22 Silicon rectifiers usually fail short-circuit, which is safe with an adequately specified transformer because it will
blow the primary-side fuse. Valve rectifiers come in two types; single cathode dual anode, which cannot fail with
one diode open circuit; and the more usual dual diode with dual filaments in parallel, which is a fire hazard and
should be replaced by a solid-state rectifier and a soft-start circuit if the equipment is to be re-certified for use. It is
a good idea to replace or re-form the smoothing capacitors when changing to solid state rectification (an increase in
output voltage occurs).
25
transformer ratio-arm bridge (TRAB) is shown below:
This circuit allows the unknown and reference impedances, the generator, and the output network,
all to have one terminal connected to ground. The transformer will introduce some minor voltage
shortfall, due to losses and leakage inductance; but it will not introduce non-linearity unless it is
operated close to core saturation, which is highly unlikely given the trivial power requirement of a
signal detection circuit. A particular disadvantage of the configuration however, is that it involves
stuffing the ratio-arm transformer with a relatively large amount of wire, which might make it
difficult to minimise stray capacitance between the primary and secondary windings. The ratio-arm
transformer must be designed with balance as the principal criterion, and it is not necessarily a good
idea to try to make it perform additional functions.
With regard to diode VRM , note that the bi-phase rectifier is simply two half-wave rectifiers feeding
the same smoothing capacitor. Hence the inverse voltage rating for a diode must be at least 22
times the RMS voltage at the transformer output.
Note that the bi-phase circuit is a frequency-doubling rectifier, which can be either an advantage or
a disadvantage in RF applications.
As mentioned before, the bi-phase rectifier gives an improvement in linearity over the half-wave
rectifier by dividing the rectifier current between two diodes; but as will be discussed in the next
and later sections, the diode forward voltage drop varies logarithmically with current in such a way
that the improvement will not be particularly large, and so the additional complexity might not be
warranted.
26
6.1
When m = 1 and Rds = 0 , the diode forward characteristic reduces to the Shockley ideal diode
equation:
23 The theory of p-n junctions in semiconductors and p-n junction transistors. W Shockley. Bell System Tech.
Journal 28(3), 1949, p435-489. See p454, equation 3.13 [Ideal diode equation with ohmic resistance].
24 Carrier generation and recombination in P-N junctions and P-N junction characteristics, C T Sah, R N Noyce,
W Shockley, Proc. IRE, Sept. 1957, p1228 - 1243. [Deviation from ideality].
25 Physics of semiconductor devices. S M Sze. Wiley, 1969. SBN 471 84290 7 [2nd edition 1981 and 3rd edition 2006
also exist]. Chapter 3, part 4 (p96-102). Current-voltage characteristics of PN junction diodes.
26 The art of electronics, P Horowitz and W Hill. 2nd edition 1989, Cambridge Univ. Press. ISBN 0-521-37095-7.
Ebers-Moll model for transistors. p79-80.
27 Most of the original articles use n for the emission coefficient, but here we will reserve n for other purposes.
27
Vd = ( kB T/qe ) ln[ ( Id / IS ) +1 ]
which, using the notation exp(x) = ex , is more usually written:
Id = IS [ exp( Vd qe / kB T ) - 1 ]
i.e.,
Id = IS [ exp( Vd / VT ) - 1 ]
The Shockley equation was derived for a P-N junction, but also applies to point-contact and
semiconductor-on-metal (Schottky or 'hot carrier' 28 ) diodes when IS and m are chosen
appropriately.
For an overview of the theory of metal-semiconductor junctions, see the review by Rhoderick29. He
points out some limitations of the near-ideal diode equation close to the threshold of forward
conduction, particularly in that it should strictly take the form:
Id = IS exp( Vd / mVT ) { 1 - exp( - Vd / VT ) }
This does not quite turn into the accepted form of the modified Shockley equation because there is
no emission coefficient (m) in the second term. It makes no difference for large diode currents, but
it does imply a small finite conduction threshold.
Bear in mind that the effective diode forward voltage drop under dynamic (i.e. AC) conditions, is
greater than the static voltage drop. That issue is examined in section 12. The static characteristic
is nevertheless an indicator of dynamic behaviour, and so governs the choice of diode. Also note
that the diode equation as given does not describe the reverse-breakdown (avalanche) region of the
characteristic; i.e., the diode is assumed to be operating within its specified VRM limit.
28
293.16 K (i.e., 20C), and we call IS at this temperature IS293 , then we can write:
ln( IS / IS293 ) = aT + b
. . . . . . . (6.2)
where T is the temperature, a is the gradient of the line, and b sets the point at which the line crosses
the axis. Now note that when IS = IS293 , ln(IS / IS293) = 0 . Thus:
0 = 293a + b
i.e.:
b = -293a
Also note that if the saturation leakage current doubles for a 20 K rise in temperature, then:
ln(2) = 313a + b = 313a - 293a = 20a
i.e.:
a = ln(2) / 20
Feeding the expressions for a and b back into equation (6.2) gives:
ln( IS / IS293 ) = T [ln(2) / 20 ] - 293 [ln(2) / 20 ]
i.e.:
ln( IS / IS293 ) = ln(2) ( T - 293 ) / 20
Exponentiating both sides and rearranging then gives:
IS = IS293 exp{ ln(2) ( T - 293 ) / 20 }
6.3
Note that this expression contains the assumption that IS doubles for a 20 K temperature rise. It
will suffice if no other information is available; but if two measurements of IS are made at different
temperatures ( T0 and T1 say), then a more accurate form can be written:
IS = IST0 exp{ ln( IST1 / IST0 ) ( T - T0 ) / ( T1 - T0 ) }
6.4
The large temperature variation of the diode characteristic, although quantifiable, suggests that
detector diodes used for precision voltage measurement applications should ideally be operated in a
temperature controlled environment. The detector, for example, might be placed inside a small
temperature regulated chamber, similar to (or actually) a crystal oven.
29
6.3 Diode measurements
The graph below shows the measured forward voltage vs. current characteristics of a variety of
small signal diodes.
The 1N4148 is a silicon P-N junction diode. The 1N5711 (Agilent 5082-2800) is a silicon Schottky
diode (i.e., silicon-metal junction diode). The OA47 is an archaic germanium gold-bonded diode,
and the rest are germanium point-contact diodes. Two 1N4148 diodes from different manufacturers
were measured merely to illustrate the point that silicon P-N diodes have the highest forward
voltage and are therefore a poor choice for low voltage detectors. The 1N5711 curve is the average
of results from four diodes, all from the same batch, which had practically identical characteristics.
The OA47 curve is the average for four diodes from two manufacturers, all having similar
characteristics. The OA90, OA91, and 1N60 curves are from single examples, and are therefore not
necessarily representative of the type. All data were recorded at an ambient temperature of 21C.
The data indicate that the IN4148, the 1N5711, and the OA47, all obey a logarithmic V/I
relationship reasonably well, whereas the germanium point-contact characteristics show
considerable curvature due to high ohmic resistance. With regard to the forward voltage drop
however, the germanium diodes are all superior to the IN5711 in the 1 to 100 A range, and the
preference for the latter may merely reflect the fact that many semiconductor manufacturers no
longer fabricate germanium. Silicon Schottky diodes, such as the 1N5711 and 1N6263 have better
high-frequency performance than germanium diodes, but germanium diodes work well at VHF and
are therefore often adequate for HF applications. Among the germanium diodes, there is little
difference between the gold-bonded and standard varieties in the 1 A to 100 A range, but the
OA47 is the best choice for currents up to 1 mA. We should observe however, that detector diodes
only conduct significantly on the peak of the applied waveform, and so the instantaneous current is
30
much higher than the average current, the difference being about an order of magnitude31 (see
section 12). Therefore, in selecting diodes for average currents in the region of 1 A to 100 A, we
should consider the steady-state voltage drop in the region 10 A to 1 mA; in which case the
germanium gold-bonded diode offers the lowest forward-drop without contest. Note however, that
one of the consequences of the diode equation is that low forward voltage-drop is associated with
high reverse saturation leakage current. Also, there will be a non-saturable reverse leakage current,
roughly equivalent to a resistance in parallel with the diode, and germanium diodes are worse than
modern silicon diodes in that respect. Hence, if reverse leakage is an issue, silicon Schottky diodes
are to be preferred. To put this matter in perspective however, the reverse leakage current of an
OA47 diode was measured as follows:
Vr / Volts
1.0
Ir / A
1.1
T = 21C
2.0
1.3
3.0
1.4
5.0
2.1
10.0
4.1
15.0
6.3
20.0
8.8
The leakage current is approximately linear in the 5 V - 20 V range and can be modelled by
assuming a parallel resistance of about 2 M. Such a defect has little effect on the operation of a
detector loaded with a 10 k - 100 k resistance, but will be deleterious in the high-impedance
detector circuits discussed earlier. Some users of Ge diodes in high-impedance detectors actually
use them without a DC return path (using the diode leakage for that purpose); but such circuits are
unpredictably dependent on the characteristics of the particular diode used.
Some final points in favour of the silicon Schottky diodes are that germanium diodes show a wider
spread of characteristics, and that the OA47 and the later AAY-series Ge-Au diodes are obsolete.
Hence the Si Schottky diodes are definitely preferable in applications requiring diode matching,
precise calibration, or availability through normal commercial channels. The 1N5711 in particular
also, has a high reverse breakdown voltage for a silicon device of its class, its VRM of 70 V making
it suitable for half-wave detectors of up to 24.7 V DC output. An OA47 half-wave detector has a
maximum DC output of 10.6 V if VRM is not to be exceeded.
31
Type
Description
1N4148
1N5711
1N5712
1N6263
AA119
AAY30
AAY32
AAY33
AAZ15
AAZ17
BAT81
BAT82
BAT83
OA47
OA90
OA91
OA95
PAD5
Si P-N junction (4 pF @ 0 V) 75
Si Schottky (2.0 pF)
70
Si Schottky (1.2 pF)
20
Si Schottky (2.2 pF)
60
Ge point contact *
45
Ge Au-bonded. High speed
30
Ge Au-bonded. High speed
30
Ge Au-bonded. High speed
12
Ge Au-bonded. High voltage 100
Ge Au-bonded. Gen. purpose 75
40
Si Schottky (1.6 pF)
50
60
Ge Au-bonded. Gen. purpose 30
Ge point contact *
30
Ge point contact
115
Ge point contact
115
Si PN low leakage (0.5 pF)
45
200
50
100
400
150
240
250
250
15
35
-
Typ Vf @ If
Typ Ir @ Vr
/V
/ mA / A
/V
max
1.0
10 0.025 20
0.41max
1
0.2
50
max
0.55
1
0.15
16
0.41max
1
0.2
50
2.6
30
170
45
0.88
150
8.0
30
0.60max
30
11
30
max
0.5
30
15
12
0.8
250
16
100
0.8
250
16
75
30
0.33
0.41
1.0
150
45
150
150
-
10
50
50
-
0.54
2.0
2.1
1.85
0.8
Data
a
b
b
f
c, d
c, d
c, d
c, d
c, d
c, d
0.1
1
15
0.2
30
30
30
30
30
5
10
300
75
80
5 pA
30
30
100
100
20
c, d
c, d
c, d
c, d
e
* Very high reverse leakage current. Generally best avoided, but leaky diodes will work in high
impedance detector circuits with no external DC return32. Not for precision measurement
applications.
Data sources
a) Philips Components Quick Reference Guide 1990.
b) Agilent 1N5711, 1N5712, 5082-2300 Series, 5082-2800 Series, 5082-2900 Schottky Barrier
Diodes for General Purpose Applications.
c) Mullard Semiconductors, 1974/5
d) Mullard Industrial Semiconductors, Quick Reference Guide 1969/70
32 Hi Fi Detector for AM broadcast. Robert Batey. http://www.g3ynh.info/circuits/hi-fi_am.html. See. TRF circuit
with regeneration.
32
e) Linear systems datasheet. www.linearsystems.com/
f) ST Microelectronics datasheet www.st.com/
Typical reverse leakage current at a given voltage gives a way of estimating the effective parallel
resistance. Provided that the reverse saturation leakage current is much smaller than the total
reverse leakage current:
Rdp = Vr / Ir
If the saturation leakage current is high, it should be subtracted from the total to find the ohmic
contribution, i.e.;
Rdp = Vr / (Ir - IS )
Point contact junctions
Note that point-contact diodes are sometimes Schottky diodes and sometimes not. If the pointcontact results in a semiconductor-metal junction, then the diode might be reasonably classified as a
Schottky type. Often however, there is an initial burn-in step in manufacture, which causes some of
the metal to diffuse into the semiconductor crystal and form a P-N junction33. Given this ambiguity,
the term 'Schottky diode' is usually reserved to mean diodes having a semiconductor-metal junction
formed by evaporation of metal on to the crystal surface.
33 Schottky barrier devices. J C Irvin and N C Vanderwal, in Microwave semiconductor devices and their circuit
applications, H A Watson (Ed.), McGraw-Hill 1969, LCCN 68-17197. p340-369, see p341.
33
34
If we put several identical diodes in series, then the forward voltage drop for a given current is
multiplied by the number of diodes (nd say). Thus
Vnd = nd Id Rds + nd m VT ln[ ( Id / IS ) +1 ]
Thus the effect is to multiply the thermal voltage (VT ) and the series resistance by the number of
diodes. The parallel resistance is, of course, similarly multiplied, as is the series partial inductance.
The capacitance however is halved.
9. Back diodes
One further rectifying device that we should mention in passing is the backward diode or 'backdiode' 34 35. This is a type of tunnel diode or Esaki diode. Due to heavy doping of the
semiconducting material (often germanium), the tunnel diode has a negative resistance region in its
V/I characteristic, which makes it useful as an amplifier, oscillator or trigger device. The back
diode is a less-heavily doped version, which means that the negative resistance characteristic largely
suppressed. This gives it a characteristic more like that of a normal diode, except that it has an
extremely low forward voltage drop, somewhere in the region of 90 mV for 10 mA forward current.
Unfortunately however, there is no general workaround as far as the diode equation is concerned.
The back diode has an extremely low reverse breakdown voltage, and a leakage current of around
1 mA for only 500 mV of reverse bias36. The reason is that, relative to the normal electrode naming
for a given material type, the back diode is used backwards; i.e., it conducts more in the 'reverse'
direction than in the 'forward' direction. This is due to the quantum-mechanical tunnelling effect
after which this class of devices is named.
The diode can therefore only be used in extremely low-impedance low-voltage circuits, which
makes it unsuitable for the construction of conventional peak detectors. It does however make a
good square-law detector, and can therefore be used for direct conversion of voltage readings into
power readings37. It also has low capacitance, and is free from charge storage effects, making some
types useful up to about 40 GHz.
Back diodes are still a current microwave technology38, but they are expensive. When a squarelaw detector is required for more general RF applications, and presuming that active circuitry can be
used, a more cost effective solution is to use a linear detector followed by an anti-logarithmic
amplifier or an analogue multiplier. Another solution is to digitise the output of a linear detector
and perform the arithmetic using a computer or a microcontroller. Detector linearisation will be
discussed in part 2.
34 The art of electronics. 2nd ed. Horowitz & Hill (previously cited). Back diode as a square-law detector, p891-892.
Tunnel diode p14-15, p1060.
35 Amplifier Handbook. Ed. R F Shea. Mc Graw Hill 1966. ISBN 0-07-056503-1. Ch. 12. Tunnel diodes and
backward diodes. C S Kim and J J Tiemann. Theory, characteristics and applications.
36 Effective noise reducer and hearing protector. R L Rod, QST, April 1978, p40. Simple audio clipper circuit
using back diodes. Gives V/I characteristic for General Electric BD1.
37 New approach to measuring SWR at high frequencies. U L Rohde. Ham Radio. May 1979, p34-35.
The Rohde & Schwarz NAUS 80 RF power meter uses back diodes as square-law detectors.
38 http://www.aeroflex.com/ams/metelics/micro-metelics-prods-TD-MTD.cfm
35
39 Radio Engineering, F E Terman, McGraw-Hill, 3rd edition 1947. Section 5.5, Diodes - Space-charge effects.
36
40 Quip usually attributed to J R Pierce of Bell Labs., but attributed by Pierce to his colleague Myron Glass.
http://www.smecc.org/john_r__pierce____electron_tubes.htm (accessed 27th Aug. 2014).
41 Electrical Indicator. T A Edison, 1884 (filed 1883), US Pat. No. 307031.
42 Inventing the history of an invention: J A Fleming's route to the valve. Sungook Hong, 2000. Artifacts, Vol 2,
02. http://www.artefactsconsortium.org/Publications/PDFfiles/Vol2Elect/2.01.ElectronicsHong,FlemingValveGr75ppiWEBF.pdf (accessed 27th Aug. 2014).
43 Improvements in instruments for detecting and measuring alternating currents. J A Fleming. 1904, British Pat.
No. 24850..
44 Instrument for converting alternating electric currents into continuous currents. J A Fleming. 1905 US Pat.
No. 803684.
37
effrontery to invent the triode45.
Being the first practical electronic amplifying device, the thermionic triode really did herald the
start of the vacuum-tube revolution46. It had two serious problems however; the first being that it
had been invented by one of Marconi's rivals, and the second being that it was superficially similar
to Fleming's valve. This led to years of litigation, and despite a 1915 disclaimer withdrawing
invalid claims in Fleming's US patent; de Forest sustained sufficient financial damage to cause him
to allow his British triode patent to lapse. That meant that the Marconi company could manufacture
and sell triode valves freely in the UK, while the US market remained in a state of legal uncertainty
until the Fleming patent expired in 1922. In 1943, the US Supreme Court declared Fleming's patent
to have been entirely invalid for all time, due to deliberate false claims and failure to declare prior
art47. That was too late to be of any use to the US electronics industry; and although it might have
entitled de Forest to some redress, nothing came of it.
According to de Forest, the Fleming valve was nothing more than a laboratory curiosity. This is
a reasonable assessment in the context of signal detection, although thermionic diodes are useful as
noise sources. The publicity surrounding the legal battles relating to it however gave it a perceived
merit that it does not actually posses, and this will have had an influence on circuit designers. The
main reason why vacuum diode detectors were used however lies in the cosy relationship that grew
up between the valve manufacturers and the equipment manufacturers, especially in the consumer
sector. Often the valve makers and the set makers were one and the same; and in Britain they
operated a vast cartel48, with restrictive practices that included approved component types and
sanctions against companies that tried to source parts independently. Hence a preference for the
valve diode detector; although it must be understood that it was only of use in circuits having plenty
of pre-amplification, such as superheterodyne radio receivers. Receivers needing sensitive lownoise detectors and mixers had to use either the triode valve, or the semiconductor crystal diode.
To be fair to the light-bulb cartel companies, it took a long time to perfect the manufacture of the
crystal detector; and a couple of spare thermionic diodes could easily be included in the superhet
final IF amplifier or first audio valve. That does not however explain the marketing of signal diode
valves (like the 6AL5 shown earlier, which was introduced in 1945) at a time when high-inverse
germanium diodes were available; or the curious practice of supplying radio sets to the British
market with valve power supply rectifiers and detector diodes, and equivalent models for the North
American market with semiconductor devices in those positions.
Meanwhile, in a parallel universe, the elements of the true electronics revolution were beginning to
come together. The unilateral conduction of crystal-metal junctions was discovered by Ferdinand
Braun49 in 1874. The use of this effect for the detection of radio signals was then developed by
Jagadis Chandra Bose from 1894 onwards50. Bose experimented with several types of
semiconductor crystal, plotting the V-I characteristics using a galvanometer capable of registering
1 nA, and appears to have anticipated the existence of P and N-type materials. He also gave public
demonstrations of radio signalling in 1895, pre-dating Marconi by two years.
45 Device for amplifying feeble electrical currents, Lee de Forest, 1907 (filed 1906), US Pat. No. 841387.
46 Although, it must be understood that de Forest's 'Audion' was too gassy to work properly as a linear amplifier, and
constitutes the invention of the control-electrode rather than a fully developed technology.
47 The genesis of the thermionic valve. Lecture given to the IEE on the 50th adversary of the invention of the
thermionic valve. G W O Howe, 1954. [ Howe's lecture was not what his audience would have expected. Refer to
the full transcript ( http://www.g3ynh.info/valves/old/history/Howe1955_genesis.pdf), not the heavily abridged
summary appearing in JIEE, March 1955, p158 ].
48 Report on the supply of electronic valves and cathode ray tubes, The Monopolies and Restrictive Practices
Commission, HMSO 1956. http://webarchive.nationalarchives.gov.uk/+/http://www.competitioncommission.org.uk/rep_pub/reports/1950_1959/020cathode.htm
49 http://en.wikipedia.org/wiki/Cat's-whisker_detector
50 The work of Jagadis Chandra Bose, 100 years of mm-wave research. D T Emerson. IEEE Microwave
Symposium (MTT-S) Digest 1997, Vol. 2, p553-556.
38
Bose's preferred detector was the galena51-metal junction, now recognised to be a type of
Schottky barrier diode. To make a diode, the semiconductor crystal requires an area contact and a
point contact, and those elements are the basis of the modern diode symbol. The springy sharpened
wire used for the point contact is known colloquially as a 'cat's whisker'. Bose used the pointcontact rectifier, in conjunction with waveguide and horn antennas, at frequencies up to 60 GHz.
He was awarded a patent on the galena diode in 1904, having filed the specification in 190152. Also,
in 1906, a patent for the silicon point-contact diode was awarded to GW Pickard53.
It is thus evident that semiconductor diodes were well known at the time when Fleming made his
'discovery', and they were considerably more sensitive than his valve even at the long-wave
frequencies that Marconi was using. Both Fleming and Marconi were also known to keep a close
watch on the patent and electrical literature. It is therefore inconceivable that they did not know of
Bose's work, and we are left to wonder why they ignored it. A possible explanation is that Fleming
was not so much looking for a detector, as looking for a novel detector that could be offered to
Marconi; and that Marconi was primarily interested in inventions he could control. When
commercial production of valves began in 1919, Marconi joined forces with GEC, a patent pooling
arrangement was agreed, and the Marconi-Osram Valve company was formed. The Marconi
company did make crystal-set radio receivers from about 1915 onwards54 55, there being a
considerable market for them; but the large valve-orientated electronics manufacturing concerns
made no self-driven effort to turn crystal diodes into stable reproducible components until after the
Second World War.
The modern semiconductor electronics revolution is closely associated with research into UHF and
microwave radar and communications, with much of the work taking place at Bell Laboratories in
the USA. There is some doubt about whether Bell Labs was first to produce a working
transistor56 57, but the company certainly made a concerted effort to investigate and perfect the
detector diode from about 1934 onwards58. About 100 different
crystalline materials were investigated by the Bell scientists, and
silicon and iron pyrites59 (fool's gold, FeS2) were found to give
the best results. Subsequent research led to the production of
uniformly active rectifying surfaces on silicon crystals,
eliminating the need to search for hot spots. This resulted in the
first manufacturable modern diode, the 1N21 front-end mixer
(used in 3 GHz radar sets) in 1942. The 10 GHz 1N23B
(introduced in 1944) is similar in appearance, and is shown on
the right.
51 Lead(II) sulphide, PbS. Crystals are cubic or octahedral, with a black opaque shiny appearance.
52 Detector for electrical disturbances. J C Bose, US Pat. No. 755840. 1904 (filed 1901).
53 Means for receiving intelligence communicated by electric waves. G W Pickard. US Pat. No. 836531, 1906
(filed 1906).
54 http://www.sparkmuseum.com/MARCONI.HTM
http://www.sparkmuseum.com/CRYSTAL4.HTM (Accessed 17th Aug. 2014)
55 The Cat's Whisker. 50 Years of wireless design. Jonathan Hill.
Oresko Books, 1978. ISBN 0-905368-47-9. p43-44.
56 Method and apparatus for controlling electric currents. J E Lilienfeld, US Pat. No. 1745175, 1930 (filed 1926).
Amplifier for electric currents. J E Lilienfeld, US Pat. No. 1877140, 1932 (filed 1928).
Device for controlling electric current. J E Lilienfeld, US Pat. No. 1900018, 1933 (filed 1928).
57 The Other Transistor: early history of the metal-oxide-semiconductor field-effect transistor, R G Arns.
Engineering Sci. and Education. Journal. Oct. 1998. p233-240.
58 Development of silicon crystal rectifiers for microwave radar receivers, J H Scaff and R S Ohl, Bell System
Technical Journal, Jan. 1947, 26(1), p1-30. http://www3.alcatel-lucent.com/bstj/
59 Vacuum-tube and crystal rectifiers as galvanometers and voltmeters at ultra-high frequencies. Arnold
Peterson. General Radio Experimenter, May 1945.
http://www.ietlabs.com/genrad/experimenters/ (accessed 17th Aug. 2014)
39
Early silicon crystal specimens incidentally were categorised into two distinct types; P-type and
N-type. P-type materials gave maximum conduction with the body of the crystal as anode (i.e,
positive with respect to the point contact) and for N-type
materials it was the other way around. This is illustrated
in the diagram on the right, and it will be seen that
modern circuit symbol is based on the N-type diode. The
Bell researchers found that N-type crystals made the best
microwave mixers; while the P-types gave better
sensitivity at low voltages, which made them useful in test equipment. Nowadays, of course, we
know that the two different types arise because of impurities; which can be removed by zonerefining of long monocrystalline bars, and reintroduced by doping of thin wafers; so that the
material resistivity and the preferred majority carrier (holes or electrons) can be controlled.
Shown here on the right is a 1950s vintage
commercial germanium diode (GEC GEX34),
fused into a 4.7mm diameter glass tube; and since
the cathode is marked with red paint, it can be
deduced that the material is N-type. In N-type
materials, the majority carriers are electrons,
which have twice the intrinsic mobility of holes.
The germanium diode was originally developed as a high-inverse-voltage signal rectifier60
(VRM > 100 V), suitable for use as the video (final) detector in radar receivers. Its advantages over
the vacuum-tube diode are nicely summarised in volume 15 of the MIT Radiation Laboratory
series61:
Much greater forward conductance.
The I-V characteristic passes through the origin (practically no contact potential).
It becomes approximately linear at low voltage.
It has very low inter-electrode capacitance, and low capacitance to ground.
It has no heater, and so produces no mains hum.
It is about the same size as a small resistor, and it doesn't need a socket.
Prior to the development of zone refining techniques, early silicon diodes had high reverse leakage;
and so the germanium diode was the first unambiguous direct replacement for the valve. Early
commercial units made their debut in post-war TV receivers62, since the 38 MHz final IF (inherited
from radar practice, and necessitated by the need for wide signal bandwidth) was fairly near to the
limit of valve diode capability.
Scientists working at Bell labs on microwave communication, and British scientists working on
wartime airborne radar development, both saw the importance of semiconductor detectors early in
their research programmes. Other radar researchers took a little more convincing however. In
1940, Britain desperately needed to develop its centimetric radar capabilities, but lacked the
manufacturing capacity required. A decision was made therefore, that Britain's radar secrets would
be shared with the USA, in return for joint R&D effort and the supply of manufactured sets. The
result was the Tizard mission in August and September 1940, during which the cavity magnetron
was revealed. That led to the establishment of the Radiation Laboratory (Rad. Lab.) at MIT in
60 The origin of semiconductor research at Purdue.
http://www.physics.purdue.edu/about_us/history/semi_conductor_research.shtml (Accessed 17th Aug. 2014)
61 Crystal Rectifiers, H C Torrey and C A Whitmer, MIT Rad. Lab. Series Vol. 15, 1948. Ch. 12, p361. High-inversevoltage rectifiers.
62 See for example, Mullard germanium diodes, Mullard Outlook, January 1954.
https://sites.google.com/site/transistorhistory2/home/philips-mullard-resources (Accessed 16 th Aug. 2014).
40
October of that year. American progress with the magnetron and its ancillary components was
rapid; and in July 1941 there was an informal showdown between British and American 3 cm radar
sets63 at Leeson House, near Swanage, in Dorset. These were bench-top tests of prototype units
intended for airborne use; and involved picking-up targets looking out across the bay towards the
Isle of Wight. This was the point at which the conventional vacuum tube was to be forever
banished from the microwave receiver front-end.
The British and American sets gave similar performance, with the small anomaly that the British
unit was the overall winner, while the Rad. Lab. set produced considerably more transmitter power.
Furthermore, tests in the US had shown earlier American receivers to be superior to the British
ones, and there appeared to have been a performance downturn. Finally a British receiver unit was
connected to an American set; and the effective detection range increased by about a factor of three.
It was quickly discovered that the discrepancy was due to the Rad. Lab. receiver having a
grounded-grid triode in the front end; whereas the British, having adopted semiconductor first
mixers at the very beginning of their microwave development programme, were using silicon
diodes. It turned out that the Rad. Lab. researchers had tried crystal mixers but abandoned them
because they felt that vacuum tubes were superior. Somewhat later it was determined that their
comparison test had been conducted using only a single diode, which had happened to be defective.
*
*
1A3 (CV753, DA90) Signal diode64
This miniature 7-pin (B7G) tube was introduced by
RCA in 1943. It probably represents one of the last
attempts to compete at VHF / UHF with the new
germanium diodes about to come on to the market.
The anode is about 2 mm tall, the anode-to-cathode
capacitance is 0.6 pF, and the self-resonance
frequency is about 1 GHz. The device still has a
heater though; and a spurious heater-to-cathode
capacitance of 0.7 pF, which inevitably restricts its
application.
With the heater connected to a 1.5 V Lithium
cell, the contact potential for the diode shown was
0.1 V when measured using a 10 M DVM.
63 The invention that changed the world. The story of radar from war to peace. Robert Buderi. Abacus, 1996.
ISBN 0 349 11068 9. see p117.
64 Technical data obtained from National Valve Museum, http://www.r-type.org/exhib/aaa0872.htm and
http://www.r-type.org/exhib/aad0031.htm (accessed 17th Aug. 2014).
41
Id = IS exp
Vp Sin - Vm
mVT
-1
(12.1)
42
input. The average diode current, Iav , and the measurable output, Vm , are the solutions to the
problem we are presently investigating.
Equation (12.1) allows us to plot the instantaneous current for a given output voltage, but we need
to determine what that output voltage is. To do that we note that:
Vm = Iav RD
The average diode current is, of course, the area under the curve for instantaneous current (Id )
divided by the length of a cycle (i.e., 2 radians). Thus:
Vm
Iav =
=
RD
Id d
It will be helpful to carry out a rearrangement and define some composite parameters in order to
perform this integration. Firstly, the exponential in (12.1) can be separated into two parts, thus:
Id = I S
exp
( ) (
-Vm
Vp Sin
exp
mVT
mVT
-1
Now notice that, because we have made the smoothing capacitor CD very large, Vm does not vary
over the course of a cycle; and so the first exponential is a constant in the integration. Vp / (mVT )
is also a constant, and so let us define:
Km = exp{ -Vm / mVT }
and
u = Vp / mVT
[ Km exp(u Sin) - 1 ] d
(12.2)
An obvious but somewhat long-winded way of performing this integration67 is to expand the
exponential as a series using:
exp(x) = e =
i=0
xi
(note that i is used here to represent a real integer)
i!
67 There is a more elegant solution to this problem involving Bessel's integral form of J 0(x). We will use it to check the
result. The longer method given here requires no special knowledge, and identifies the error in Weller's article.
43
ui Sini
Iav =
K
-1
m
2
i!
0
i=0
IS
(12.3)
General expressions for Sini are given by Ryshik and Gradstein68. When a sine is raised to an odd
power, the generating function is:
1
Sin2n-1 =
2n-2
n-1
R&G 1.320.3
k=0
and when a sine is raised to an even power, the generating function is:
n-1
Sin2n =
(-1)n-k .2. 2nCk Cos{ 2 (n-k) } + 2nCn
22n
k=0
R&G 1.320.1
where qCr is a binomial coefficient (pronounced "q choose r" ), defined as:
q!
q
Cr =
q(q-1) . . . . . (q-r+1)
=
r! (q-r)!
r!
R&G, p433
Dwight, p1
Expansions of Sini for powers up to 7 are given by Dwight69. Evaluation of binomial coefficients
is assisted by noting that they are the elements of Pascal's triangle. In this case however, it also
pays to note the following:
In the expansion of Sin2n-1 (1.320.3), every term has the sine of an integer multiple of as a
factor.
In the expansion of Sin2n (1.320.1), the terms have the cosine of an integer multiple of as a
factor, except for a single constant ( 2nCn / 22n ).
Sinusoids (i.e., sines and cosines) that undergo an integer number of cycles over the range of an
integration average to zero.
The corollary is that all of the sine and cosine terms from these expansions vanish in the integration
process, leaving only the constants from the even power series. To make use of that information,
we need to change the summation index in (12.3) because we want a series that only produces even
powers. The substitution required is i = 2n. Putting that into (12.3) gives:
68 Tables of Series, Products, and Integrals. I M Ryshik and I S Gradstein. VEB Deutscher Verlag der
Wissenschaften, Berlin, 1957. p26 and 27. 1.320.1 (even), 1.320.3 (odd).
69 Tables of Integrals and Other Mathematical Data. H B Dwight. 4th edn. Macmillan 1961. LCCN: 61-6419. Page
82. Formulae 404.12 - 404.17.
44
u2n . 2nCn
Iav =
K
-1
m
2n
2
(2n)! 2
0
n=0
IS
u2n (2n)!
Iav =
K
-1
m
2
(2n)! 22n n! (2n-n)!
0
n=0
IS
(u/2)2n
Iav =
K
-1
m
2
(n!)2
0
n=0
IS
Now recall that u = Vp / (mVT ) is a constant in the integration. Therefore everything in the integral
is now a constant, and the indefinite integration process is simply a matter of multiplying the whole
thing by . Then in the upper integration limit we set =2, and the the lower limit we set =0.
Subtracting lower limit value of the integral (=0) from the upper limit value gives:
(u /2)2n
Iav = IS
K
-1
m
(n!)2
n=0
An infinite series with the square of a factorial in the denominator is the classic signature of a
Bessel function. Reference to McLachlan71 or Dwight72 reveals the series to be the modified Bessel
function73 of the first kind, zero order, I0(u).
I0(u) = J0(ju) =
(u /2)2n
n=0
where j = -1
(n!)2
. . . . . . . . . . . . (12.4)
There is, incidentally, a well known integral form for the ordinary Bessel function J0(x) that
70 Notice here that 2n-n = n(2-1) = n , so that (2n-n)! = n! . This is where the mistake in Weller's article occurred.
He did not give his working, and wrote 2! instead of n! .
71 Bessel Functions for Engineers. N W McLachlan. 2nd edition. Oxford, Clarendon Press 1955. p200. eqn. 153.
72 Dwight (already cited), p195, 813.1.
73 Modified Bessel functions are the functions that result from evaluating an ordinary Bessel function using a purely
imaginary argument.
45
corroborates the result just given74 75 :
2
1
J0(x) =
2
exp(jx Sin) d
1
I0(u) =
2
exp(u Sin) d
= IS exp
( ) ( )
-Vm
Vp
I0
mVT
mVT
-1
(12.5)
A solution for the rectified output voltage (Vm ) can be obtained from this expression. Note
however that there are two instances of Vm , one separable and one as an exponent. This means that
there will be no closed-form analytical solution; although finding a solution by trial and error is
fairly straightforward.
More instructively however, we can compare the expression with the modified Shockley
equation as given at the beginning of this section:
Id = IS [ exp{ Vd / mVT } - 1 ]
Here, Vd can be taken to be either the instantaneous diode voltage drop under dynamic conditions,
or the diode forward voltage drop under static conditions (i.e., the error that occurs if we place it in
series with a DC voltmeter). So let us define the effective diode forward voltage drop under
dynamic conditions as:
Vf = Vp - Vm
This is, of course, the error that occurs when the diode is used as a peak detector.
In order to extract Vf from equation (12.5), we can define a new function (W0 say) that provides a
factor that converts an exponential into a modified Bessel function. This gives:
46
Vm
Iav =
RD
= IS W0 exp
( ) ( )
Vp
-Vm
exp
mVT
mVT
-1
(12.6)
= IS W0 exp
( )
Vf
mVT
-1
(12.6a)
( )
Vf
+ 1 = W0 exp
IS R D
mVT
Now taking the natural logarithm of both sides and rearranging, we get:
Vf = m VT ln
Vm
+1
IS R D
- ln(W0)
(12.7)
( )
Vp
I0
( )
Vp
= W0 exp
mVT
mVT
Thus:
I0( Vp / mVT )
W0 =
exp( Vp / mVT )
Inserting this into (12.7), and noting that -ln(W0) = ln(1/W0) , gives:
Vf = m VT ln
Vm
+1
IS R D
) (
exp( Vp / mVT )
+ ln
I0( Vp / mVT )
(12.8)
Now bearing in mind that: Vm = Vp - Vf , this expression can be solved manually by entering an
initial guess for Vp - Vf and adjusting it until the two instances of Vf agree. Experimentation is
also facilitated by noting that Bessel functions are built-in to modern spreadsheets (such as Open
Office Calc). Program routines to perform this iteration process automatically must be carefully
47
designed however, there being a problem of reliable convergence when Vp is close to zero (this
matter is resolved in section 13.3).
If we put equation (12.8) in terms of the average diode current we get:
Vf = m VT ln
Iav
+1
IS
) (
exp( Vp / mVT )
+ ln
I0( Vp / mVT )
(12.9)
The first term is identical to the DC form of the diode equation except that it has an average current
instead of a constant current. The second term is therefore a correction for the fact that the true
current is varying over time. Thus we can write the effective diode forward voltage drop under
dynamic conditions as:
Vf = Vf _ + Vf ~
where the component associated with the continuous output current (DC) is:
Vf _ = m VT ln
Iav
+1
IS
(12.9dc)
exp( Vp / mVT )
I0( Vp / mVT )
(12.9ac)
12
12 32
12 32 52
I0(u) =
1+
+
+
+......
2
3
(2 u)
1! 8u 2! (8u)
3! (8u)
exp(u)
u 10
(12.10)
48
The polynomial in square brackets is not given in compact form in the books cited, but if we give it
the symbol PaI0(u) (polynomial used in the asymptotic form of I0 ), it can be written:
PaI0(u) = 1 +
n=1
(2n-1)!
(n-1)! 2n-1 (8u)n
as u
PaI0(u) 1
(12.10p)
The asymptotic form of I0(u) is good to 6 significant figures for u 10 , and can be evaluated to 5
significant figures with only 4 terms in the polynomial (i.e., terms up to n = 3). To put that in
perspective with regard to the diode voltmeter problem, note that if mVT = 29.1 mV (1N5711 @
20C), then the asymptotic form applies when the peak input voltage is 0.29 V. For a silicon
Schottky detector diode, the input threshold voltage for a usable output (and a starting point for
accurate non-linearity correction) is somewhere around 0.35 V. Thus we find that the asymptotic
form of the Bessel function is applicable even at the nominal threshold of usability of the detector.
The wide-range of applicability of the asymptotic form should not be taken to imply that we can
disregard the general analytical form; particularly because the general form can (within its
limitations) extend the range of our diode voltmeter theory to zero peak input. It does however
allow us to characterise the large-signal limiting behaviour of the model with good precision using
simple formulae.
If we substitute (12.10) into (12.9ac), with the polynomial written as PaI0(u), we get:
Vf ~ = mVT ln{ (2 u) / PaI0(u) }
. . . . . . . . . . . . . . . . . (12.10ac)
where u = Vp / mVT
This can be separated into three parts, with the square root eliminated by halving the logarithm:
Vf ~ = ()mVT ln(2) + ()mVT ln(u) - mVT ln{ PaI0(u) }
The positive terms and the total are shown
plotted on the right for mVT = 29.1 mV. The
term involving the log of the polynomial is
included, but it is not shown on the graph
because it makes a contribution of only
-0.38 mV when Vp is 0.29 V, and -10 V when
Vp is 10 V (see spreadsheet det_models.ods,
sheet 5). The constant term mVT ln(2) is
26.7 mV, and the total varies between 60 mV
and 112 mV as the peak input changes from 0.3
V to 10 V.
The error that would be incurred by
computing the forward voltage drop using
average current in place of direct current (i.e., by
ignoring the dynamic correction) is fairly small
for large input; around 1% for inputs of about 10 V peak and a 1 M load. For precision absolute
voltage measurement however, it is evident that the correction should be included.
49
The issue of how to calculate the transfer function of a diode detector using the theory just
developed is the subject of section 13 (calculation procedures). Some representative graphs,
modelling the behaviour of a 1N5711 detector with a 100 k load resistance are shown below.
These were produced using the function program DVfp2m( RD ; Vp ; mVT ; IS ), which is described
in section 13.3 (see spreadsheet det_models.ods, sheet 3, for the working details).
The graph on the right shows the diode
dynamic forrward voltage-drop over a
detector peak input voltage range from
1 mV to 10V. Also shown are the separate
DC and AC contributions, the latter having
been calculated using equation(12.9ac).
Notice that the AC contribution is very
different from the DC characteristic. This
means that linearity compensation schemes
using reference diodes in DC circuits
cannot provide complete correction.
50
12.2 Peak to average current ratio
An expression for the instantaneous diode current was given earlier as equation (12.1)
Id = IS exp
Vp Sin - Vm
mVT
-1
By inspecting this we can see that the peak forward current occurs when Sin =1 , i.e., when
= +90 or /2 radians. Thus we can easily write a separate expression for the peak forward
current (which we will call Ip ) :
Ip = IS exp
Vp - Vm
mVT
-1
But Vp -Vm is the effective diode forward drop under dynamic conditions, to which we previously
assigned the symbol Vf . This we have:
Ip = IS [ exp( Vf / mVT ) - 1 ]
which can be rearranged:
( Ip / IS ) + 1 = exp( Vf / mVT )
. . . . . . . . . . . . . . . . (12.11)
Also, somewhat less trivially, we obtained an expression for the average diode current by
integrating the expression for instantaneous current. This was given by equation (12.6a) as:
Iav = IS [ W0 exp( Vf / mVT ) - 1 ]
which can be rearranged:
( Iav / IS ) + 1 = W0 exp( Vf / mVT )
. . . . . . . . . . . . . . . . (12.12)
1
=
( Iav / IS ) + 1
exp( Vp / mVT )
=
W0
I0( Vp / mVT )
51
Ip + I S
exp( Vp / mVT )
=
Iav + IS
(12.13)
I0( Vp / mVT )
Now, bearing in mind that the reverse saturation leakage current (IS ) is small (a few nA for a silicon
Schottky diode), this quantity is effectively the peak to average current ratio. It is also the ratio that
governs the AC correction term in the effective diode forward voltage drop.
The error in the detector output ( Vf = Vp - Vm ) was given by equation (12.9) as:
Vf = mVT ln
Iav
+1
IS
+ mVT ln
Vf _
exp( Vp / mVT )
I0( Vp / mVT )
Vf ~
This solution has given us the basis for calculating the error, and it tells us that the AC contribution
is dependent only on the peak input voltage. That conclusion is somewhat opaque however,
because it does not seem to offer a physical reason for the effect. All becomes clear when we use
(12.13) as a substitution:
Vf = mVT ln
Iav
+1
IS
( )
Ip + IS
+ mVT ln
(12.14)
Iav + IS
Now it can be seen that the AC error is governed by the peak to average current ratio; which is
physically reasonable, and perhaps also obvious in retrospect.
A corollary of equation (12.14) is that the AC error disappears when the peak current is the same as
the average current. This has a trivial meaning when both are zero; but it also implies that the error
will disappear when the rate of change of the detector input voltage is so slow that the output can
follow it. In this simplified derivation, we have prevented that from happening by specifying that
the smoothing capacitor must be 'large'. This gives rise to a further and perhaps more interesting
corollary; which is that the dynamic voltage error is independent of frequency provided that the
capacitor is large. That point leads to a useful conclusion, as will now be explained.
It is possible to make extremely linear active detectors ('superdiode' circuits) by the use of high
gain and negative feedback. Such detectors however have limited bandwidth. We can nevertheless
calibrate a diode voltmeter against an active detector by comparing the readings when both are
driven simultaneously by a signal generator working within the active detector's range.
Furthermore, we can then vary the output voltage of the generator until the output of the diode
detector agrees with that of an identical detector connected to an (unknown) RF voltage that we
want to measure. The level adjustment of the reference generator can of course be done
automatically. The reading given by the linear detector is then a measurement of the unknown RF
voltage. This type of voltmeter is known as an 'amplitude tracking detector'. We will revisit the
idea in part 2.
52
12.3 Detector power dissipation and input impedance
In section 12.1, for the purpose of deriving the basic transfer function, it was assumed that the
generator driving the detector has zero output impedance. This is unrealistic, of course, but we can
adapt the model to deal with finite source impedance provided that we know the detector input
impedance at the driving frequency. The actual input voltage is obtained from a potential divider
formed by the source impedance and the input impedance (see section 1.6).
A rough idea of the AC input resistance of a practical half-wave rectifier circuit was given in
section 1.5. Here we will carry out a more detailed analysis. The determination is a matter of
applying the principle of conservation of energy, so that the power dissipated in the detector and its
DC network is equal to that abstracted from the source.
The instantaneous power delivered to the detector is the product of the instantaneous input
voltage, multiplied by the instantaneous input current. Thus:
Pinst = Vin Id = Vp Id Sin
Where, as was originally given as equation (12.1):
Id = IS exp
Vp Sin - Vm
mVT
-1
and
u = Vp / mVT
Then, assigning the symbol Pdet to the average power (i.e., the power):
V p IS
Pdet =
2
Although this integral is not the same as in the expression for average current given in section 12.1,
the discussion in that section is nevertheless relevant to its solution. Recall (as mentioned in the
text following equation 12.3) that all pure sinusoids average to zero over a complete cycle. Thus, if
we were to multiply-out the part of the expression above that is to be integrated, we would end up
with a term -Sin, which averages to 0. Hence, the integral reduces to:
53
V p IS K m
Pdet =
2
Now, as was previously done in order to solve the integral equation (12.2), we can expand the
exponential as an infinite series. In this case however, we also need to multiply every term in the
series by Sin, which has the effect of increasing the index for Sini by 1. Thus:
Vp IS Km
Pdet =
2
ui Sini+1
d
i!
0 i=0
As was also discussed previously, raising a sinusoid to an integer power results in a series of
harmonics, plus a constant in those cases where the power is even. All of the harmonics, being
sines and cosines of integer multiples of , vanish in the integration, leaving only the series of
constants. Since the integral then contains no instances of , integration between 0 and 2 is
reduced to a matter of multiplying the whole thing by 2 (which cancels the existing factor of 1/2).
The series of constants is given by 2nCn / 22n , where 2nCn is a binomial coefficient (see the
discussion following 12.3), and in this case, 2n = i+1 . To produce only the even terms, we can
change the summation index to n and use the substitution: i = 2n-1 . Note that for i=0 , i+1 is
odd, so the new series starts from n=1. Thus:
Pdet = Vp IS Km
n=1
u2n-1
2n
Cn
(2n-1)! 22n
The binomial coefficient is evaluated as: 2nCn = (2n)! / (n!)2 . As before, the square of a factorial in
the denominator should immediately alert us to the possibility that the series represents a Bessel
function. Also, since the expression for the current involves a zero-order modified Bessel function
of the first kind, and the matter of getting from a current to a power, in general, involves raising the
order of an expression; we might reasonably be able to guess which Bessel function it is. Thus the
solution in this case is a matter of proving the most obvious conjecture. Expanding the binomial
coefficient gives:
Pdet = Vp IS Km
n=1
u2n-1 (2n)!
(2n-1)! 22n (n!)2
This expression can be simplified using the substitutions: (2n)! = (2n-1)!2n and 22n = 222n-1 .
The result is:
54
Pdet = Vp IS Km
(u/2)2n-1 n
(n!)2
n=1
The series summation (as expected) corresponds to the first-order modified Bessel function of the
first kind 78, I1(u) . Thus the solution for the total detector power, averaged over a complete cycle
of the input waveform is:
Pdet = Vp IS Km I1(u)
After expansion of the composite parameters Km and u, this becomes:
( ) ( )
-Vm
Pdet = Vp IS exp
Vp
I1
mVT
(12.15)
mVT
Now, if RZin represents the resistive component of the detector input impedance due to AC to DC
conversion; then, since the power delivered to the detector (excluding losses due to causes other
than conversion) is, by Joule's law:
Pdet = Vin(RMS)2 / RZin
and, for a sinusoidal input waveform:
Vin(RMS) = Vp / 2
Then
RZin = Vp 2 / 2 Pdet
12.16
Note however, that (12.15) is not the best method for calculating the detector power because it will
run into the floating-point arithmetic upper range limit when Vp / mVT is large. An unrestricted
approach will be developed in the next two sections.
55
12.4 Diode power dissipation
The preceding derivation provides the basic solution to the detector input impedance problem; but
in the matter of providing insight into the behaviour of the detector, it is somewhat unsatisfactory.
In particular, it should be obvious that, for moderately large signal levels, the power dissipation in
the detector is dominated by the dissipation in the load resistance, which is trivially obtainable as
the product of the output voltage Vm and the average diode current Iav . The interesting information
is that which governs the non-linear relationship between input voltage and input impedance, this
being related to the dissipation that occurs in the diode. The diode power can be written:
Pdiode = Pdet - Pload = Pdet - Vm Iav
Now recall that in section 12.1 , for the purpose of extracting the diode dynamic forward voltage,
Vf , we defined a function W0(Vp /mVT) , which converts an exponential into a zero-order modified
Bessel function of the first kind. Here, we will perform a similar trick by defining a function
W1(Vp /mVT) , which converts an exponential into a first-order modified Bessel function of the first
kind. Equation (12.15) then becomes:
( )
-Vm
Pdet = Vp IS exp
( )
Vp
W1 exp
mVT
mVT
( )
Vf
Pdet = Vp IS W1 exp
(12.17)
mVT
Iav = IS W0 exp
( )
Vf
mVT
-1
( )
Vf
Pdiode = Vp IS W1 exp
mVT
- Vm IS W0 exp
( )
Vf
mVT
-1
56
Pdiode = IS exp
( )
Vf
mVT
( Vp W1 - Vm W0 ) + Vm
(12.18)
( ) (
Vf
exp
Iav
mVT
+1
IS
1
W0
I1( Vp / mVT )
Pdiode = ( Iav + IS ) Vp
- Vm
I0( Vp / mVT )
+ IS V m
(12.19)
This is the general expression for diode power, but it should also be noted that there exists an
asymptotic form79 80 of I1(u) , which (like the asymptotic form of I0(u), equation 12.10) is applicable
over the entire practical working range of an uncorrected detector (i.e., Vp /mVT > 10). The
asymptotic form can be written:
I1(u ) exp(u) PaI1(u) / (2 u)
. . . . . . . . . . . . . (12.20)
(4-12) (4-32)
-
2! (8u)2
1! 8u
-......
3! (8u)3
as u
PaI1(u) 1
(12.20p)
or alternatively:
PaI1(u) = 1 +
n=1
(-1)n
n! (8u)
4 - (2i-1)2
i=1
(12.20s)
57
The asymptotic forms of the zero and first order modified Bessel functions are identical except for
the polynomial used, which means that if we substitute them into (12.19) we end up with the ratio
of the two polynomials, i.e.;
PaI1( Vp / mVT )
Pdiode = ( Iav + IS ) Vp
- Vm
PaI0( Vp / mVT )
+ IS Vm
Vp /mVT > 10
12.21
This is an extraordinary result because, although we might adopt the habit of imagining that Vf is a
DC voltage present in the detector network; it is actually a transitory voltage (the peak detection
error), which only exists for an instant in each cycle at the point when Sin=1. Still, for
moderately large inputs, it is apparent that we can estimate the diode power on the basis that it is an
average voltage to be multiplied by the average diode current.
58
A graph of diode power vs. input voltage for a
detector with a 1 M load, mVT = 29.1 mV and
Is = 2.2 nA is shown on the right. The curve
marked 'Exact' is generated by equation (12.19),
and the curve for for the approximation Iav Vf
(equation 12.21) lies a little above it. The overestimation occurs because the ratio of Bessel
functions included in (12.19) has the effect of
reducing the effective value of Vp slightly. The
error is around 4% for reasonably large inputs
(see spreadsheet det_models.ods, sheet 3), but
since the diode power is a small proportion of the
total in that case, the effect on input impedance
estimation is fairly minor (see section 12.6).
In section 12.1 it was shown that Vf can be split into two components, Vf _ (DC) and Vf ~ (AC).
This separation also leads to two power components, which are shown on the graph. The curve for
Iav Vf _ shows that neglecting the dynamic (AC) component leads to a considerable
underestimation of diode power.
I1( Vp / mVT )
Pdet = ( Iav + IS ) Vp
- Vm
I0( Vp / mVT )
+ IS Vm + Iav Vm
12.22
I0( Vp / mVT )
Routines for the calculation of the Bessel functions I1 and I0 and their asymptotic forms are
developed in section 13 (calculation procedures). In subsection 13.9 however, the various methods
are combined into a single program that calculates the ratio. The program chooses initially between
the small argument and asymptotic forms and then computes both the zero and first order functions
from within the same program loop. This approach is probably about as efficient as it gets within a
given programming environment, and is, of course, free from practical argument-range restrictions.
The detector power is then given by:
Pdet = ( Iav + IS ) Vp RatioI1_0( Vp /mVT )
where RatioI1_0( ) is the function Combining this result with equation (12.16) gives the input
59
impedance:
RZin = Vp 2 / 2 Pdet
i.e.:
Vp
I0( Vp / mVT )
2 ( Iav + IS )
I1( Vp / mVT )
RZin =
12.23
. . . . . (12.24)
60
Vp2 ( 1 - Vf / Vp ) / RD Vp 2 / 2 RZin
i.e.:
RZin RD / [ 2 ( 1 - Vf / Vp ) ]
12.25
Rdiode Vf / Iav
12.26
and
Iav = (Vp - Vf ) / RD
Hence:
Rdiode RD Vf / (Vp - Vf )
i.e.:
Rdiode RD / ( Vp / Vf - 1 )
12.27
The exact expression for the diode dynamic resistance is obtained by rearranging (12.26):
Rdiode = 2 RZin - RD
12.28
RZin being obtained from (12.23). This final expression is however probably only of use for the
purpose of evaluating approximations; since the point of calculating Rdiode is to determine the input
impedance, and so there is no need to do it if the input impedance is known.
61
Vf = m VT ln
Vm
+1
IS R D
) (
exp( Vp / mVT )
+ ln
I0( Vp / mVT )
Vm = Vp - Vf
Where Vm is the DC output (measurement) voltage, Vp is the peak input voltage, and Vf is the
effective diode forward voltage drop under dynamic conditions.
The expression above is however tautologous, in that any substitution to reduce the three
principal variables to two (e.g., by replacing Vm with Vp - Vf ) results in the need to input a value of
a voltage that we want to find before it will give an output of that quantity. This means that there
are no closed-form analytical solutions. Nevertheless, it is a straightforward matter to set up the
various parts of the equation in the columns of a spreadsheet, and then input trial solutions until
both instances of the desired quantity agree. That method is, of course, tedious and unsatisfactory;
and so the problem is that of how to automate the process (i.e., turn it into a set of algorithms). We
will moreover want solutions in either direction, i.e.; so that we can put in a value of Vp and obtain
Vf and Vm ; or so that we can put in a value of Vm and obtain Vf and Vp . There is also a particular
difficulty in this case, which is that the argument of a logarithm function-call must never become
less than zero; and so for small values of Vm we will be performing an iterative procedure close to a
point at which a runtime error will occur.
A further issue is that the the problem involves exponentials and modified Bessel functions of
large argument. In Open Office, for example, the double-precision floating-point range is from
4.94110-324 to 1.79810308, and a simulation for a 1N5711 detector operating at 20C (for
example) hits the upper limit at Vp = 20.62 V. The built-in spreadsheet function for I0 is moreover,
not accessible from the macro programming environment. It is therefore useful to have routines for
the modified Bessel function and its asymptotic form.
62
I0(x) =
n=0
(x /2)2n
(n!)2
This has a simple recursion factor, which can be spotted by writing down the first few terms:
term(0) = 1
term(1) = (x/2)2 / ( 12 )
term(2) = (x/2)2 . (x/2)2 / (12 . 22 )
term(3) = (x/2)2 . (x/2)2 . (x/2)2 / ( 12 . 22 . 32 )
etc.
Hence the overall pattern is:
term(n) = term(n-1) (x/2)2 / ( n n )
Furthermore, the quantity (x/2)2 can be pre-calculated as x x /4 .
The resulting algorithm, coded in Open Office Basic, is shown below:
Function I0Bessel(byval x as double)as double
'Calculates modified Bessel function, 1st kind, zero order, I0(x)
I0Bessel = 1
If x <= 0 then exit function
Dim xx4 as double, term as double, sum as double, rcf as double
Dim n as integer
term = 1
sum = term
xx4 = x*x/4
n=0
do
n = n+1
rcf = xx4/(n*n)
term = term*rcf
sum = sum + term
loop until term/sum < 1E-12
I0Bessel = sum
end function
This routine works for arguments up to 708, as does the built-in spreadsheet function; and it agrees
with the built-in function to at least 12 significant figures (see spreadsheet file det_models.ods,
sheets 1 & 2).
63
13.2 Polynomial used in the asymptotic form, first kind, zero order
For arguments greater than about 10, the asymptotic form of the modified Bessel function I0(x) can
be used. It was given earlier as equation (12.10)
12
12 32
12 32 52
I0(x) =
1+
+
+
+......
(2 x)
1! 8x 2! (8x)2
3! (8x)3
exp(x)
x 10
This form will not extend the argument range of the function itself because it is the output, rather
than some internal variable, that falls outside the floating-point arithmetic range. We can however,
use it in the AC part of the expression for Vf , in which case it eliminates the exponential and vastly
increases the calculation range.
The AC part of Vf , originally given as equation (12.9ac), is:
Vf ~ = m VT ln
exp( Vp / mVT )
I0( Vp / mVT )
12 32
12 32 52
+
1! 8x
+
2! (8x)
+......
3! (8x)
64
Function PaI0(byval x as double)as double
'Calculates the polynomial used in the asymptotic form of I0(x)
PaI0 = 1
If x < 9.6 then exit function
'floating-point underflow occurs for arguments < 9.6
Dim x8 as double, term as double, sum as double, rcr as double
Dim n as integer
term = 1
sum = term
x8 = 1/(8*x)
n=0
do
n = n+1
rcr = x8*(2*n-1)*(2*n-1)/n
term = term*rcr
sum = sum + term
loop until term/sum < 1E-9
PaI0 = sum
end function
A comparison of calculations of Vf ~ using the small argument formula and the polynomial formula
is given in the spreadsheet file det_models.ods, sheet 5. The two expressions agree to within a
picovolt for arguments > 60 ; except that the small argument formula, as mentioned earlier, fails for
arguments > 708 ; whereas the polynomial version was tested initially with arguments up to 35481
without error. The latter corresponds to just over 1030 V peak input for the 1N5711 example, not
that a small signal diode can survive this.
Vf = Vf ~ + mVT ln
Vp - Vf
+1
IS R D
In order to solve this equation for Vf , we can start by giving different symbols to the two instances,
i.e.:
y = Vf ~ + mVT ln
Vp - x
+1
IS R D
and x = Vf when x = y
(13.1)
If we set-up this equation in a spreadsheet, it is a simple matter to insert a value of x into the right-
65
hand side and see what value of y comes out. The input x can then be adjusted until it agrees
with y to a number of significant figures deemed sufficient, and then x (or y) is the solution. The
point however, is to find a procedure that will work automatically and never fail; regardless of the
peak input voltage Vp ; the parameters mVT and IS RD ; and the initial, intermediate and final
values of x.
A starting point for guessing x is to note that, by definition, it can never be smaller than Vf ~ .
Hence it is safe to begin by setting x = Vf ~ . The simplest automatic strategy for finding the
solution is then to make x equal to the value of y that comes out, and try again. The process is
repeated until x and y converge.
In fact, in a test with mVT = 29.1 mV and IS RD = 2.2 mV, the simple strategy worked well for
input voltages (Vp ) above about 200 mV. For smaller inputs however, it failed miserably due to
overshoot. Feeding back the first value of y caused Vp - x to be < -1. That caused the argument of
the log function to be < 0, resulting in a program error. Furthermore, the reason for developing the
program is to explore the behaviour of the detector in the poorly-characterised region of the diode
forward-conduction threshold.
In order to make the procedure stable for small inputs, we need to find a more accurate way of
estimating the required value of x. To do that, we can analyse the adjustment process as follows:
When x x + x ,
y y + y
66
successful iteration by about an order of magnitude in comparison to the simple procedure. There is
however, still a tendency to overshoot for very small inputs, this being due to the inaccuracy of the
derivative in the first round of iteration. The problem is easily solved however, by adding an extra
nested program loop. In this loop, x is added to x, and the argument of the logarithm is tested to
see if it is greater than 1. If the argument is too small, the original value of x is restored, x is
halved, and the program returns to the point at which x is added to x. Thus x is successively
reduced until the log argument is in range.
The complete algorithm is shown below. It has been tested for values of Vp down to 1 pV (well
outside any practical measurement range), at which point it is still both stable and convergent. Note
that Is is called Isat in the program, because 'is' is a Basic keyword.
Function DVfp2m(RD as double, Vp as double, mVT as double, Isat as double) as double
'calculates voltage error (forward drop) of simple diode peak detector.
'Calls the functions I0Bessel( ) and PaI0( ).
DVfp2m = 0
if Vp <= 0 then exit function
Dim u as double, Vfac as double, x as double, y as double, Vsr as double
Dim kd as double, arg as double, der as double, deltax as double, xold as double
'Calculate Vf~. Use the asymptotic form for arguments >= 60.
u = Vp/mVT
if u < 60 then
Vfac = mVT*log( exp(u) / I0Bessel(u) )
else
Vfac = mVT*log( sqr(2*pi*u) / PaI0(u) )
end if
'Set up starting values and composite parameters
x = Vfac
Vsr = Isat*RD
arg = 1 + (Vp-x)/Vsr
kd = -mVT/Vsr
'Calculate the difference between x and y, the derivative, and the estimated shift in x
do
y = mVT*log(arg) + Vfac
diff = x-y
der = kd/arg
deltax = diff/(der-1)
'Apply shift to x and check that the log argument is valid. If not, reduce the shift and try again.
do
xold = x
x = x + deltax
arg = 1 + (Vp-x)/Vsr
if arg <=1 then
deltax = deltax/2
x = xold
end if
loop until arg > 1
loop until abs(diff) < 1E-12
DVfp2m = x
end function
67
13.4 Determining peak input voltage from output voltage
The computational methods described so far are useful for modelling detectors; but for practical
measurement purposes, the matter of interest is that of inferring the AC input voltage (Vp ) from the
DC output. This problem is made difficult by the fact that we cannot pre-calculate the the AC
component of the diode forward voltage ( Vf ~ ), because we need to know Vp in order to do it.
Also, the unknown is used in the argument of a modified Bessel function, and the choice of
calculation method depends on that argument.
Small voltages
To solve the peak input problem for small voltages, it is probably easiest not to use equation (12.8)
but to go back to its precursor, equation (12.5):
Vm
Iav =
RD
= IS exp
( ) ( )
-Vm
Vp
I0
mVT
mVT
-1
This can be rearranged into a form that would be analytical but for want of an inverse of the
modified Bessel function.
( ) (
Vp
I0
Vm
mVT
) ( )
Vm
+1
IS R D
exp
mVT
The right-hand side of this expression can be calculated, and if we call it y (say), then:
Vp = mVT AntiI0(y)
where AntiI0(y) is the inverse of the modified Bessel function of the first kind, zero order.
Modified Bessel functions have a single-valued inverse because, unlike the ordinary Bessel
functions, they are not undulatory. Indeed, the curves for the In (first kind) functions roughly
resemble exponential growth. Inverse calculation routines however, do not appear to be standard
library functions. That problem was solved by writing a suitable routine; which will be described in
section 13.6.
Large voltages
A limitation of the solution given above, of course, is that it contains an exponential and will easily
exceed the floating-point computation range in typical applications. To circumvent that problem,
we must find an expression that eliminates the exponential. That can be done by substituting the
asymptotic form of the modified Bessel function (12.10 - 12.10ac) into the AC part of equation
(12.8), thereby producing the expression:
Vf = m VT ln
Vm
+1
IS R D
) (
(2 x)
+ ln
PaI0(x)
x = Vp /mVT
Vm = Vp - Vf
We already have a routine for the polynomial PaI0(x) (see section 13.2), and we know that it
68
converges for arguments greater than about 10 and does not deviate greatly from 1. Therefore it can
be seen that solutions based on the expression above will be viable for x between 10 and the
floating-point limit.
Now, substituting for Vf and using x = Vp /mVT throughout we get:
Vm
x-
= ln
mVT
Vm
+1
IS R D
) (
(2 x)
+ ln
PaI0(x)
ln
Vm
+1
IS R D
Vm
+
= x - ln
mVT
(2 x)
PaI0(x)
The left-hand side can now be calculated, and if we call it (say) y , we have:
y = x - ln{ (2 x) / PaI0(x) }
The logarithm can also be separated into terms, giving:
y = x - () ln(2) - () ln(x) + ln{ PaI0(x) }
This can be solved iteratively by making an estimate of x (x1 say), which can be inserted into the
formula above to produce an approximation to y (y1 say). Also notice that y is a good first
approximation for x because the first term on the right is much larger than all of the others. The
required change in y (y) can then be used to estimate the required change in x (x) via the
derivative dy/dx, i.e.:
y = y - y1 x dy/dx
Thus:
x ( y - y1 ) / ( dy/dx )
Note that 1/(dy/dx) is not the same as dx/dy because the original relationship is non-linear. The
differential relationship does however become exact at the point at which the solution is found, and
so y - y1 0 is the termination criterion for the iteration process. The derivative is easily obtained
by applying the chain rule to the final term. Thus:
dy/dx = 1 - 1/(2x) + { dPaI0(x)/dx } / PaI0(x)
One additional mathematical function is required, and that is the derivative of the polynomial. It is
easily obtained by differentiating the series term by term, and a suitable calculation routine is given
in section 13.5. Its contribution to the overall derivative is very small and negative; but it must be
included if the calculated shift in x is to vanish at the point of solution.
69
Overall calculation
The overall calculation routine using the techniques just described is given below. Note that it
returns the detector forward voltage drop (i.e., the detector error), which can be added to the output
voltage to determine the peak input voltage.
For detector outputs of less than 1.5 V, the inverse modified Bessel function method is used, the
main calculation being performed by a separate routine called Anti_I0(). For higher voltages, the
iterative method is used, and this involves calls to calculate the polynomial and its derivative.
Procedure development, prior to coding, was carried out in the spreadsheet det_models.ods,
sheet 3. The program was tested, with IS RD = 2.2 mV and mVT = 29.1 mV, for Vm values between
1 pV and 100 kV. The changeover point at Vm = 1.5 V will not lead to out-of-range polynomial or
polynomial derivative arguments when used in conjunction with modern semiconductor diodes, but
it can easily be changed to deal with unusual diodes or large diode stacks if necessary.
70
13.5 Derivative of the asymptotic form polynomial, first kind, zero order
The polynomial used in the asymptotic (large argument) form of the zero order, first kind, modified
Bessel function (see section 13.2) is given by the series:
12
12 32
PaI0(x) = 1 +
12 32 52
+
2! (8x)2
1! 8x
+......
3! (8x)3
2 . 12 32
-
1 . 8 x2
3 . 12 32 52
-
1 . 2 . 82 x3
4 . 12 32 52 72
-
1 . 2 . 3 . 83 x4
-.....
1 . 2 . 3 . 4 . 84 x5
Thus:
Term(0) = -1/8x2
Term(1) = (-1/8x2 ) 12 32 / 1 . 8x
Term(2) = Term(1) 52 / 2 . 8x
Term (n) = Term(n-1) (2n+1)2 / 8x n
The calculation procedure is given below. Note that in order to avoid underflow errors, the
termination criterion is set to 10-9 for arguments above 13.4, and 10-6 for arguments between 9.7 and
13.4. This applies to OO Basic. Other programming environments may differ.
Function DerPaI0(byval x as double)as double
'Calculates the derivative PaI0'(x)
DerPaI0 = 0
If x < 9.7 then exit function
Dim x8 as double, term as double, sum as double, rcf as double, test as double
test = 1E-9
If x < 13.4 then test = 1E-6
term = -1/(8*x*x)
sum = term
x8 = 1/(8*x)
Dim n as integer
n=0
do
n = n+1
rcf = x8*(2*n+1)*(2*n+1)/n
term = term*rcf
sum = sum + term
loop until term/sum < test
DerPaI0 = sum
end function
71
13.6 Inverse modified Bessel function, first kind, zero order
As mentioned earlier, the modified Bessel functions all have a unique inverse; and as has been
found here, it is sometimes useful to be able to determine it. No strenuous effort has been made to
find the most efficient method for the inverse of the first kind, zero order, but the following
approach works well.
If y = I0(x) , then let us define x = AntiI0(y) and find its value iteratively.
Given y, we first need to make an estimate of x (x1 say). The estimate does not have to be
particularly accurate, provided that the iteration procedure converges from its starting point. Using
x1 we can calculate the corresponding value of y (y1 say) using the routine for I0(x) that we
already have, i.e.;
y1 = I0(x1)
We can now estimate the required shift in x using the procedure described in the previous section,
i.e.:
x ( y - y1 ) / ( dy/dx )
where dy/dx is, in this case, the first derivative of the modified Bessel function, first kind, zero
order; which just happens to be the modified Bessel function of the first kind, first order, i.e.:
dy/dx = I0'(x) = I1(x)
A routine for calculating this function will be given in the next section. x is added to x1 and a
new value of y1 is calculated. This iteration process is continued until x becomes very small.
Prior to coding, the procedure was set up for manual implementation in the spreadsheet
det_models.ods, sheet 4. It was found that for values of y up to about 10, the routine would
converge rapidly if the initial estimate for x was always set to 1. For very large values of y
however, this leads to much unnecessary calculation; and as the floating-point limit is approached it
can result in overshoot and runtime error.
An improved estimation procedure for large x values was obtained by noting that the polynomial
used in the asymptotic form of I0(x) never deviates greatly from 1. Therefore a reasonably good
estimate for large arguments is given by omitting it. Thus, for x > 10 :
y = I0(x) exp(x) / (2 x)
Unfortunately, this cannot be rearranged to give x on its own, but if we set a value of 10 to the
instance of x that occurs in the square root bracket we get:
exp(x) y (20)
i.e.:
x ln(y) + () ln(20)
It was found that for very large values of y, approaching the floating-point limit, this simple formula
estimates x to an accuracy of about 0.3%. Hence the calculated x is extremely accurate, and a
72
termination criterion of |x| < 110-12 has the iteration completed within two or three cycles. This
high accuracy holds until y is less than about 400 (x < 7.93), but the iteration procedure will still
terminate correctly if it is used down to y = 1. For smaller inputs the estimate asymptotes to
x ln(20)/2 = 2.07. Since x lies between 0 and 1.81 for y < 2, it was felt that setting x = 1 for
y < 2 would probably eliminate at least one cycle of iteration in the lower input ranges.
The resulting program routine is shown below. It is accurate to at least 12 decimal places, and in
the OO Basic programming environment, using the companion functions I0Bessel( ) and
I1Bessel( ) , it works for inputs up to 9.110305.
73
13.7 Modified Bessel function, first kind, first order
The first derivative of the modified Bessel function I0(x), as used in the inverse procedure described
above, is the same as the modified Bessel function I1(x). This first order function is also used in
calculating diode power and detector input impedance. The series form is given by Dwight82, but it
can also be obtained by differentiating the series for I0(x) (see section 13.1) term by term. The
result is:
(x/2)3
(x/2)
I0'(x) = I1(x) =
+
1
(x/2)5
+
1 2
+
2
(x/2)7
1 2 3
+...... =
2
1 2 3 4
n=1
n (x/2)2n-1
(n!)2
74
13.8 Polynomial used in the asymptotic form , first kind, first order
For arguments greater than about 10, the asymptotic form of the Bessel function I1(x) can be used.
This was given earlier as equation (12.20):
I1(x ) exp(x) PaI1(x) / (2 x)
Using this form will not extend the argument range for the Bessel function itself, because it is the
returned value that falls outside the floating-point range when the argument is large. The detector
input impedance determination however uses a ratio of Bessel functions, I1(x)/I0(x) . The two
asymptotic forms have a common factor exp(x)/(2 x) ; and so the exponentials cancel when their
ratio is taken, thereby greatly increasing the argument range. The Polynomial PaI1(x) is given by:
4-12
PaI1(x) = 1 -
(4-12) (4-32)
-
1! 8x
2! (8x)
-......
3! (8x)
Thus:
Term(0) = 1
Term(1) = (4-12) / 1 . (-8x)
Term(2) = (4-12) (4-32) / 1 . 2 . (-8x) . (-8x)
Term(3) = (4-12) (4-32) (4-52) / 1 . 2 . 3 . (-8x) . (-8x) . (-8x)
Term(n) = Term(n-1) . [ 4 - (2n-1)2 ] / n (-8x)
An OO Basic function is given below:
Function PaI1(byval x as double)as double
'Calculates the polynomial used in the asymptotic form of I1(x).
PaI1 = 1
If x < 9.6 then exit function
Dim x8 as double, term as double, sum as double, rcf as double
Dim n as integer
term = 1
sum = term
x8 = -1/(8*x)
n=0
do
n = n+1
rcf = x8*(4-(2*n-1)*(2*n-1))/n
term = term*rcf
sum = sum + term
loop until abs(term/sum) < 1E-9
PaI1 = sum
end function
75
13.9 Ratio of modified Bessel functions, first order / zero order
The ratio of modified Bessel functions I1(x)/I0(x) is used for calculating the input impedance of a
diode detector. The routine below calculates both Bessel functions in the same program loop for
arguments < 15, and calculates both asymptotic form polynomials in the same program loop for
larger arguments.
Function RatioI1_0(byval x as double) as double
'Calculates ratio of Bessel funcs; I1(x)/I0(x).
RatioI1_0 = 0
If x <= 0 then exit function
Dim term1 as double, sum1 as double, rcf1 as double,
Dim term0 as double, sum0 as double, rcf0 as double
Dim n as integer
n=0
If x < 15 then
Dim xx4 as double
term1 = x/2
term0 = 1
sum1 = term1
sum0 = term0
xx4 = x*x/4
do
n = n+1
rcf1 = xx4/( n*(n+1) )
rcf0 = xx4/(n*n)
term1 = term1*rcf1
term0 = term0*rcf0
sum1 = sum1 + term1
sum0 = sum0 + term0
loop until term1/sum1 < 1E-9 and term0/sum0 < 1E-9
else
Dim x8 as double
term1 = 1
term0 = 1
sum1 = term1
sum0 = term0
x8 = 1/(8*x)
do
n = n+1
rcf1 = -x8*(4-(2*n-1)*(2*n-1))/n
rcf0 = x8*(2*n-1)*(2*n-1)/n
term1 = term1*rcf1
term0 = term0*rcf0
sum1 = sum1 + term1
sum0 = sum0 + term0
loop until abs(term1/sum1) < 1E-9 and term0/sum0 < 1E-9
end if
RatioI1_0 = sum1/sum0
end function
76
13.10 Determining output voltage from source off-load voltage
In later sections, we will show how practical diode voltmeter problems can be transformed so that
they can be solved using techniques developed for the simple detector model. In this way, the diode
detector is turned into an absolute AC voltage measuring instrument; at least in the sense that an
accurate measurement of the DC output can be corrected to produce an accurate measurement of the
peak amplitude of a sinusoidal input. In order to do that however, it is necessary to take account of
the source loading effect (perhaps not so importantly when 50 signal generators are connected to
detectors driving CMOS op-amps; but certainly when feeble sources such as non-invasive sampling
networks are used to drive moving-coil meters). Thus, in order to complete the minimum set of
mathematical tools, we need two more procedures; one to get from source off-load voltage to DC
output (for modelling purposes); and one to get from DC output to source off-load voltage.
The problem of calculating the output
voltage from the source voltage can be
understood by considering the circuit on
the right. Here VS is the RMS off-load
source voltage, and Vin is the RMS
voltage produced by the source when it it
loaded by the detector. RS +jXS is the source output impedance, and RZin is the detector input
impedance. Thus Vin is the output of a potential divider formed by the two impedances, i.e.:
Vin = VS RZin / ( RZin + RS + jXS )
The detector however does not preserve phase information, and so we can eliminate complex
quantities from the analysis without affecting its generality. Thus, taking the magnitude of the
denominator, and adopting the convention that an RMS voltage not written in bold is a magnitude,
we get:
Vin = VS RZin / { ( RZin + RS )2 + XS2 }
. . . . . . . (13.2)
We still cannot calculate Vin however, because RZin depends on it, and so it is necessary to use an
iterative procedure. That involves making an initial guess of Vin , calculating the average current
Iav , and hence the input impedance , and using the results to refine the original guess. The relevant
formulae and functions are:
Iav = Vm / RD
where, using the function described in section 13.3:
Vm = Vp - Vf = Vp - DVfp2m( RD ; Vp ; mVT ; Is )
Vp = Vin 2
and, using equation (12.23) and the Bessel function ratio routine from section 13.9:
RZin = Vp2 / 2 Pdet = Vp / 2 (Iav + Is) RatioI1_0( Vp /mVT )
The approach was tested in the spreadsheet det_models.ods, sheet 6, and then coded into the Basic
macro function shown below. The first method tried was that of taking the value of Vp from a
round of calculation and using it directly as the starting value for the next round. This resulted in a
77
convergent procedure provided that |RS +jXS | was less than RD , but for very large magnitudes of
source impedance and source voltages in the region of 0.1 V to 1 V (when mVT = 29.1 mV ), the
routine would terminate at the maximum allowed number of iterations without giving the correct
answer. Subsequent investigation showed that the internal value of Vp in this case would oscillate
above and below the solution value, with the error neither growing nor shrinking. This behaviour
was greatly curtailed by taking the seed for the next round of iteration to be the simple average of
the previous and the current value. For some extremely improbable inputs however, such as a
source impedance magnitude of 1000RD , the routine could still fail. The complete solution was to
use the 2:1 weighted average in favour of the previous value, thus eliminating any possibility of
overshoot.
Note that the termination criterion for the program loop limits the maximum number of iterations
to 128. Unmodified, it will not fail to find a solution for realistic input parameters, but the
restriction gives insurance against 'program not responding' errors in the event that some awkward
combination of inputs is found.
The results of some simulations using this algorithm are shown in the graphs below (see
spreadsheet det_models.ods, sheet 6).
78
79
13.11 Determining source off-load voltage from output voltage
For the purpose of making absolute voltage measurements, the single most important routine is that
which allows the off-load source voltage to be determined from the detector DC output. Given the
work already done however, obtaining the source voltage is straightforward because it is quasianalytical (i.e., it is analytical except for the fact that it calls on numerical routines).
Given Vm we can calculate Vp using the function described in section 13.4:
Vp = Vm + DVfm2p( RD ; Vm ; mVT ; Is )
also:
Iav = Vm / RD
Using these quantities in equation (12.23) (with the function from section 13.9):
RZin = Vp / 2 (Iav + Is) RatioI1_0(Vp /mVT)
The on-load input voltage is the peak voltage divided by 2 :
Vin = Vp / 2
and the source off-load voltage is given by a rearrangement of equation (13.2):
VS = Vin [ { ( RZin + RS )2 + XS2 } ] / RZin
A Basic macro function that performs the calculation is shown below:
Function DetVm2s(byval Vm as double, Rs as double, Xs as double, RD as double, _
mVT as double, Isat as double) as double
'Calculates source off-load voltage from detector output voltage. Version 1.00,
'Calls functions DVfm2p( ) and RatioI1_0( )
Dim Vp as double, Vs as double, Iav as double, Rzin as double, Vin as double
Vp = Vm + DVfm2p(RD, Vm, mVT, Isat)
Iav = Vm / RD
Rzin = Vp /( 2*(Iav+Isat)*RatioI1_0(Vp/mVT) )
Vin = Vp / sqr(2)
DetVm2s = Vin*sqr((Rzin+Rs)^2 +Xs^2)/Rzin
end function
While the source off-load voltage itself is now easily obtained however, it is still necessary to
calculate its standard deviation. We will return to that problem in a later section.
80
Fig. 14.1
This quasi-general diode voltmeter model is very different from the simple model of section 12; but
if we keep sight of the practical objective (which is to determine the off-load source voltage VS
from the DC output voltage Vm ), then it should be apparent that there are various rearrangements
that will simplify the problem. We can start by considering the parasitic reactances.
Lds is the series partial inductance of the diode and its wires or circuit tracks. Its value is about
the same as that of a piece of wire of the same total length and average diameter as the diode leads
from the source terminal to the smoothing capacitor. There will also be a contribution from the
ground-return, but the use of a ground-plane will minimise this. In some circumstances, such as
when working with surface-mount devices at VHF and below, or when using wire-ended diodes
with short wires at HF, its effect will be swamped by the overall measurement uncertainty; in which
case it can be neglected. The best thing to do with it is to lump it with the source impedance. In
81
that case, when calculating the detector input voltage from the output voltage, we will get the
quantity Vin' (or its AC component), which is not strictly observable, but we can still calculate Vin~
(if there is any point in doing so explicitly) and thence VS , or we can calculate VS directly using the
modified source impedance.
Cj is the diode capacitance. This is a function of the junction depletion-layer thickness and so
varies according to the instantaneous voltage (the varactor effect), but for the purpose of rectifier
modelling, its value is taken to be the average at the excitation frequency. The fact that we are
ignoring harmonic generation due to capacitance variation is a potential source of systematic error,
but the effect is small for low capacitance signal diodes and will be lost in the overall measurement
uncertainty. Thus having declared the intention of treating Cj as a simple capacitance; we can note
that it cannot alter the average DC voltage across the smoothing capacitance CD , it consumes no
power, and since CD is intentionally vastly greater in value then Cj , the only effect that Cj has is to
place a weak reactive shunt across the AC input. Thus we might as well lump it with the source
impedance.
The equivalent circuit that results from moving the parasitic reactances is shown below. It is
now possible to combine them with the source impedance, but we will not do so yet because there is
at least one more element to be transferred. For the purpose of the transformations to follow, the
various voltages and currents associated with the diode network have also been marked.
Fig 14.2
As was demonstrated in section 12, determining the detector transfer function depends on being
able to calculate the average output current, Iav . That information is obtained by integrating the
instantaneous diode current, Id , over a complete cycle of the input waveform, but in this case the
process is complicated by the presence of the resistances Rds and Rdp . We can however, easily
solve that part of the integral involving the non-saturable leakage current Id", and thereby obtain
another transformation.
Previously, we adopted the convention that the AC component of the input waveform is
sinusoidal and is zero and positive-going at t = 0 , i.e.:
Vin~ = Vp Sin
(some commentators prefer the cosine form, but it makes no difference to the outcome). Thus we
can define the instantaneous input voltage Vin' as:
Vin' = Vp' Sin' - Vbk
where both the peak voltage and the phase angle of the AC component are primed because there
will be a shift in both magnitude and phase due to the parasitic reactances. To the right of the point
82
at which Vin' is marked on the diagram however, there are no more finite reactances, and so there
will be no more phase shifts on the path to the load.
The instantaneous voltage across the whole diode (including its series resistance) is:
Vd = Vp' Sin' - Vbk - Vm
Hence the instantaneous diode current, using the diode equation (6.1), is:
Id = Id' + Id" = IS [ exp( Vd' / mVT ) - 1] + ( Vp' Sin' - Vbk - Vm ) / Rdp
where
Vd' = Vp' Sin' - Id' Rds - Iav ( RD + Rport )
Note here that, because the expression for Vd' involves Id', there is no closed-form analytical
expression for the total instantaneous current, let alone its integral (although a numerical solution is
possible, as we will see later). Finding the average of the non-saturable leakage current is however
straightforward.
The total average current is the area under the curve for the total instantaneous current.
Vm + Vbk
Iav =
1
=
RD + Rport
1
Id d =
1
Id' d +
2
Id" d
Thus, if we separate that part of the average current that flows through Rdp and call it -Iav" (the
reason for the minus sign will become apparent shortly), we have:
1
-Iav" =
2
Now observe that the sinusoidal term averages to zero over a complete cycle. Thus the constant
terms are all that remain, and after evaluating the integral we get:
-Iav" = -( Vbk +Vm ) / Rdp
What this means is that the diode parallel resistance acts as a DC shunt across the total load
resistance RD + Rport . The minus sign in the expression above tells us that the current flowing
through Rdp is subtracted from the current available for charging the smoothing capacitor, i.e., it
reduces the detector output. This result, actually obvious on re-inspection of Fig. 14.2, allows us to
redraw the model once again.
83
Fig. 14.3
The sign chosen above for Iav" allows the diagram to be labelled sensibly. Also notice that there are
now two instances of Rdp in the equivalent circuit. The reason for that is that Rdp, as well as
shunting the rectified output, also shunts the input signal; and since we have imposed a
mathematical separation between the AC and DC parts of the problem, it is necessary to account for
both effects separately.
Having worked out how to deal with Rdp however, it should be noted that in many instances it
can be ignored. This is because modern signal diodes typically have a large Rdp ; about 250 M for
a 1N5711 Schottky diode, and somewhere from a few hundred k to a few M for a germanium
detector diode. Thus, given that most RF networks will have output impedances of less than a few
hundred ohms magnitude; the AC shunting effect will usually be negligible. The DC shunting
effect should however be taken into account when using a diode in conjunction with a large value of
RD or Rport , a sensible guideline being that it should be included if it reduces the output by more
than about 0.1%.
We have now annexed most of the diode equivalent-circuit elements either into an effective source
impedance, or into an effective load resistance, or both. A new, greatly simplified model results:
Fig. 14.4
Here the voltages Vin~' and Vin~" are the AC components of the voltages Vin' and Vin". Also notice
that the smoothing capacitance has been given a prime, but this is merely to warn that the
transformed load resistance RD' is not to be used for calculating the detector time constant.
For this model, Vm' is the total detected voltage, i.e.:
Vm' = Vm + Vbk
and
RD' = (RD + Rport ) // Rdp
Thus Vm can be calculated as the output of a potential divider composed of RD and Rport , i.e.:
84
Vm = Vm' RD / (RD + Rport )
and Vm' can be calculated from Vm :
Vm' = Vm (RD + Rport ) / RD
i.e.:
Vm' = Vm (1 + Rport / RD )
. . . . . . (14.1)
Fig. 14.5
85
This network can now be modelled as a simple diode detector, with new circuit parameters suitable
for passing to the calculation routines described in sections 13.10 and 13.11. All we have to do is
pass the modified parameters from the circuit below in place of the original quantities.
Fig. 14.6
Notice that adding Rds to the effective source impedance does not change its reactance.
Consequently, although the effective source resistance gets an extra prime (cf. fig. 14.4) the reactive
element does not.
To calculate Vm", the appropriate function call is:
Vm" = DetVs2m(VS ; RS" ; XS' ; RD" ; mVT ; IS )
and to calculate VS , the call is:
VS = DetVm2s(Vm" ; RS" ; XS' ; RD" ; mVT ; IS )
The required parameters can be calculated using the formulae and programs given below.
Effective load resistance
The effective load resistance is given by:
RD" = RD' + Rds
Where RD' is the effective load resistance from fig. 14.4. Thus:
RD" = Rds + Rdp // (RD + Rport )
i.e.:
RD" = Rds + [ Rdp (RD + Rport ) / ( Rdp + RD + Rport ) ]
14.2
86
Effective source impedance
To calculate RS" and XS', the transformation shown in the diagram below is required:
Fig. 14.7
. . . . . . . (14.7)
Extracting expressions for RS' and XS' from this is related to a general problem that we will refer to
as the 'series-parallel to series (SP2S) transformation'. Since we will also require exactly the same
derivation again later, it will be set out using a more general notation in the next subsection. The
resulting formulae are coded into Basic macros, so that RS" and XS' can be calculated using the
following function calls:
RS" = Rds + SP2S_R( RS ; XS +XLds ; Rdp ; XCj )
and
XS' = SP2S_X( RS ; XS +XLds ; Rdp ; XCj )
The actual expressions for RS" (or RS') and XS' are rather complicated and, in principle at least, we
will need to differentiate both of them with respect to each of their parameters in order to calculate
the overall standard deviation of a measured voltage. This is not inherently difficult, but it is
algebraically messy. It is therefore worth noting that Cj and Rdp are usually of considerably less
statistical importance than the other parameters. If |Xcj| and Rd are relatively large, it is still
useful to include them for the purpose of reducing systematic error, but they won't make much
difference to the estimated standard deviation (ESD) of either RS" or XS'. Therefore we can usually
neglect them in that context. Hence for the purpose of calculating ESD, we merely have to
differentiate the expressions:
RS" RS + Rds
and
XS' XS + XLds
87
Calculating DC output from total detected voltage
The total detected voltage Vm" is determined from the measured output voltage Vm by analysing the
potential divider formed by Rds , Rdp , Rport and Rd :
Rdp // (RD + Rport )
Vm + Vbk = Vm"
. . . . (14.3)
Rds + Rdp // (RD + Rport )
and
RD
Vm = ( Vm + Vbk )
. . . . (14.4)
RD + Rport
Fig. 14.8
RD
Vm = Vm"
[ Rds + Rdp // (RD + Rport ) ] ( RD + Rport )
Now if we expand the parallel resistances we get:
Rdp (RD + Rport )
Rd
( Rdp + RD + Rport )
( RD + Rport )
Vm = Vm"
Rdp (RD + Rport )
Rds +
( Rdp + RD + Rport )
Rearranging the main denominator so that all of its terms share a common denominator gives:
RD Rdp
( Rdp + RD + Rport )
Vm = Vm"
Rds ( Rdp + RD + Rport ) + Rdp ( RD + Rport )
( Rdp + RD + Rport )
Thus, with some rearrangement of the terms in the denominator:
RD Rdp
Vm = Vm"
14.5
Rds ( RD + Rport ) + Rdp ( RD + Rport + Rds )
Notice here that, as Rdp , only the terms in the denominator having Rdp as a factor survive, and
so we get:
as Rdp , Vm Vm" RD / ( RD + Rport + Rds )
88
If we decide to ignore the diode series resistance, or set it to zero so that we can use a numerical
method for the average diode current, we get:
as Rds 0 , Vm Vm" RD / ( RD + Rport )
A basic macro routine that calculates Vm from Vm" using equation (14.5) is given below:
Function DCVout( Vm2 as double, RD as double, Rport as double, Rdp as double, _
Rds as double) as double
DCVout = Vm2*RD*Rdp/(Rds*(Rd+Rport) + Rdp*(RD+Rport+Rds))
end function
The function call is:
Vm = DCVout( Vm" ; RD ; Rport ; Rdp ; Rds )
or, if starting from Vm' (i.e., using a numerical diode-with-series-resistance model):
Vm = DCVout( Vm' ; RD ; Rport ; 1 ; 0 )
Notice, by inspection of equation (14.5), that Rdp makes no difference to the calculation in this case,
but it must be > 0 . Inserting a value of 1 is therefore convenient.
89
Calculating total detected voltage from DC output
To obtain the total detected voltage Vm" from the output voltage Vm , we can rearrange equation
(14.5) thus:
Rds ( RD + Rport ) + Rdp ( RD + Rport + Rds )
Vm" = Vm
RD Rdp
This can be simplified in various ways, e.g.:
Vm" = Vm
Rport
Rds ( RD + Rport + Rdp )
1+
+
RD
RD Rdp
14.6
90
14.2 Series-parallel to series transformation
The diagram below refers to a frequently encountered electrical problem. A generator having finite
source impedance has an impedance placed across its terminals, and we want to find the new
effective source impedance.
Fig. 14.9
// jXPP
RSS + RPP + jXSS
91
-RPP XSS XPP + j RSS RPP XPP
RSP + jXSP =
( RSS RPP - XSS XPP ) +j( RPP XSS + RSS XPP + RPP XPP )
and multiplying numerator and denominator by the complex conjugate of the denominator:
(-RPP XSS XPP +jRSS RPP XPP) [( RSS RPP -XSS XPP) -j( RPP XSS +RSS XPP +RPP XPP)]
RSP + jXSP =
( RSS RPP - XSS XPP )2 + ( RPP XSS + RSS XPP + RPP XPP )2
Now we can multiply-out the numerator and separate the real and imaginary parts.
RSP + jXSP =
-RPP XSS XPP ( RSS RPP -XSS XPP) + RSS RPP XPP ( RPP XSS +RSS XPP +RPP XPP)
+j[ RSS RPP XPP ( RSS RPP -XSS XPP) + RPP XSS XPP ( RPP XSS +RSS XPP +RPP XPP) ]
( RSS RPP - XSS XPP )2 + ( RPP XSS + RSS XPP + RPP XPP )2
Thus, with a reversal of the terms in the real part to get rid of the leading minus sign:
RSS RPP XPP ( RPP XSS +RSS XPP +RPP XPP) - RPP XSS XPP ( RSS RPP -XSS XPP)
RSP =
14.8
2
( RSS RPP - XSS XPP ) + ( RPP XSS + RSS XPP + RPP XPP )
and
RSS RPP XPP ( RSS RPP -XSS XPP) + RPP XSS XPP ( RPP XSS +RSS XPP +RPP XPP)
XSP =
14.9
2
( RSS RPP - XSS XPP ) + ( RPP XSS + RSS XPP + RPP XPP )
92
14.3 Shunt diode rectifier with port impedance and parasitics
The problem of transforming the shunt-diode rectifier circuit so that it can be analysed using the
simple detector model has much in common that of the series-diode case. There are however, some
differences that must be taken into account.
A fairly realistic equivalent circuit for the shunt-diode detector is shown below. As is the
convention in this document, a voltage placed in square brackets refers to the DC component of a
waveform. The diode model used is as discussed in section 8.
Fig 14.10
Notice that the source impedance has been written with an additional subscript 0 on each of its
elements. This is done pending a transformation to make the complete network analytically
equivalent to the series diode half-wave detector. Additionally, a perfect coupling capacitor has
been included to make it explicit that the source is a DC open-circuit. This might seem
unnecessary, given that there will be an actual coupling capacitor, but defining the series reactive
element of the source impedance as negative at the input frequency is not the same as stating that
there is no DC path.
The major analytical difference between the conventional detector and the shunt detector is that
the latter has one or more port impedances, instead of just a port resistance. This is because a port
resistance is defined (at least in this document) as the DC resistance in the path to the load; and in
the case of the shunt detector, that path is physically separate from the source impedance. Thus we
have to take both the total port resistance and the total port impedance into account.
The reason why there are two port impedances is that the shunt detector can be used for the
measurement of floating voltages. Thus, assuming the grounding arrangement shown in the
diagram, Zport2 is the ground-return impedance, whereas Zport1 refers to the network connecting the
diode to the live output terminal. When the source AC reference is not at ground potential, it can
still usually be arranged that |Zport2| is very small relative to |Zport1| . In that case most analyses can
proceed with Zport2 = 0 , but it is important to be aware that its neglect might introduce a systematic
error.
The total port impedance for analytical purposes is:
Zport = Zport1 + Zport2
If the maximum possible detector sensitivity is required, Zport1 can be the impedance of an RF
choke; although such a choice is not recommended for precision voltmeter applications. Even if the
DC path is provided by a resistor however, it is still an impedance, not a pure resistance, because
high-value resistors are usually predominantly capacitive at radio frequencies when operated below
the SRF.
93
The next step in the analysis is to separate the port resistance from the total port impedance. This is
illustrated in the diagram below.
Fig 14.11
In this representation, a perfect choke channels the DC output through the port resistance, and the
port impedance is made into a DC open circuit by means of a perfect coupling capacitor. Note that
the port resistance is not the same as the resistive element of the port impedance. For that reason
the impedance has been written as RP // jXP . The parallel form is used because the transformation
has made it obvious that the port impedance is effectively in parallel with the detector input. This is
so because the smoothing capacitor intentionally has a reactance that is very much smaller in
magnitude than |RP // jXP| , and the AC component of the current through the smoothing capacitor
makes no contribution to the DC output. Hence the next step in the transformation, as shown
below, is to place the port impedance directly in parallel with the source.
Fig 14.12
At this point it will become apparent why the source impedance has been given symbols that are
different from the conventional detector case. The reason is that, while the port impedance is,
notionally, already combined with the source impedance in the series diode network, in this case we
need to combine the two impedances explicitly. Thus, for the effective source impedance (i.e.,
including the shunting effect of the port impedance), we can use the same definition as in the
conventional case (and hence the same computer programs for enumeration) provided that we make
the following transformation:
RS + jXS = ( RS0 + jXS0 ) // RP // jXP
Here, the use of the parallel form for the port impedance works to our advantage because, in the
preferred case when the port impedance is that of a high-value resistor; the resistor can usually be
represented as its actual resistance in parallel with a capacitance of about 0.4 pF (provided that it is
94
operating below its SRF). Furthermore, if the source impedance magnitude is relatively low, the
effect of the port resistance on the AC amplitude can be neglected, and the port impedance becomes
simply the reactance of the resistor's capacitance. Note incidentally, that the port capacitance as a
modelling parameter will be strongly correlated with the junction capacitance Cj . Thus, if the
model is used for parameter extraction, the two capacitances will probably have to be accounted for
by eliminating one and allowing the other to increase slightly.
The matter of combining the source impedance and the port impedance requires the seriesparallel to series (SP2S) transformation described in section 14.2. Using the Basic macro routines
given in that section, the required function calls are:
RS = SP2S_R( RS0 ; XS0 ; RP ; XP )
and
The diode parasitic reactances and parallel resistance can now be combined with the source
impedance, and another instance of the parallel resistance can be combined with the DC load. This,
as is shown below, gives us a configuration that can be analysed using the diode-with-seriesresistance numerical method.
Fig 14.13
This network should be compared with figure 14.3, whereupon it will be noted that; if the source
and port impedances are combined, and the diode with its series resistance and the ideal choke are
transposed, the only difference lies in the ordering of the resistances in the DC potential divider.
Since addition of resistances is commutative, this problem is analytically identical to the
comparable series diode case.
Finally, instances of the diode series resistance can be combined with the source impedance and the
load.
Fig 14.14
Once the port impedance is combined with the source impedance, this network is analytically
identical to the one shown in figure 14.5. It can therefore be analysed as a simple detector using the
transformations given in the previous section.
95
96
15.1 Instantaneous diode current
In the diagram below, the simple detector model of section 12 has been modified by the inclusion of
diode series resistance. In order to find the average diode current Iav , from which the DC output
voltage Vm is calculated, it is necessary to integrate the instantaneous current Id over a complete
cycle of the input waveform. The first task therefore is to determine Id . This is given by the near
ideal diode equation (see section 6):
Id = IS [ exp{ Vd' / mVT } - 1 ]
Where:
Vd' = Vd - Id Rds
i.e.:
Id = IS [ exp{ ( Vd - Id Rds ) / mVT } - 1 ]
The problem with this expression is that it has instances of Id both inside and outside the
exponential bracket. This means that there will be no closed-form analytical solution. We will
therefore use an iterative method, based on the procedure first introduced in section 13.3. We start
by giving different symbols to the two instances of Id :
y = IS [ exp{ ( Vd - x Rds ) / mVT } - 1 ]
An initial value for x is easily obtained by calculating Id on the assumption that Rds= 0 , i.e.;
x0 = IS [ exp( Vd / mVT ) - 1 ]
x then has to be adjusted until y agrees with it to some sensible degree of accuracy, i.e:
|x - y| maximum acceptable error.
The change in y due to a small change in x is defined thus:
When x x + x ,
y y + y
97
x ( x - y ) / ( y/x - 1)
The derivative is obtained using the chain rule:
y/x = IS (-Rds / mVT ) exp{ ( Vd - x Rds ) / mVT }
Notice that if Rds = 0, the derivative is always zero and the initial estimate for x is the solution.
A Basic program that performs the iteration is shown below. Algorithm development was carried
out in the spreadsheet det_models.ods, sheet 2. The convergence criterion |x-y| has been set so
that Id is calculated to the nearest pico amp.
Function DinstI(byval Vd as double, Rds as double, mVT as double, Isat as double) as double
'Calculates diode instantaneous current when diode has finite series resistance
Dim x as double
x = Isat*(exp(Vd/mVT)-1)
DinstI = x
If Rds <= 0 then exit function
Dim arg as double, y as double, deriv as double, diff as double, deltax as double
do
arg = (Vd - x*Rds)/mVT
y = Isat*(exp(arg)-1)
deriv = -Isat*Rds*exp(arg)/mVT
diff = x-y
deltax = diff/(deriv-1)
x = x+deltax
Loop until abs(diff) < 1E-12
DinstI = x
end function
98
15.2 Average diode current by numerical integration
The DC output of a detector is given by the product of the average diode current and the effective
load resistance; i.e., assuming zero port resistance and zero non-saturable leakage current:
Vm = Iav RD
The average current is the area under the curve for the instantaneous current. This can be obtained
by integrating the instantaneous current ( Id ) over one complete input cycle and dividing by the
length of a cycle, i.e.;
Vm
Iav =
1
=
RD
Id d
This was carried out in section 12.1 for the case with Rds = 0 and led to a solution involving the
zero-order modified Bessel function of the first kind. We know immediately however, from the
working in the previous section, that when Rds > 0, this integral is not analytical because Id is not
analytical. A numerical integration method is therefore required. Such procedures are generally
less efficient than analytical methods, and so before applying them, it is sensible to look for ways in
which computational redundancy can be eliminated.
Throughout this document, we have made the assumption that the reactance of the smoothing
capacitor is very small (i.e., analytically zero) at the excitation frequency. This is actually not an
approximation in the determination of the DC output for a given detector input, because RF currents
in the smoothing capacitor cannot affect the average (recall that sinusoids average to zero over an
integer number of cycles), but it will affect the detector input impedance slightly. Still, this effect,
essentially an adjustment of the distribution of harmonic currents, is very small in comparison to the
issue we are trying to resolve; which is
whether or not it is safe to neglect the
dissipation of harmonic energy in the diode
series resistance. Therefore we will retain the
large-smoothing-capacitance approximation,
which has the effect of eliminating outputvoltage droop in the interval between the
diode-current pulses.
If the DC output voltage does not droop,
then the current-spike that occurs at the peak
of the input voltage waveform will be
symmetric about the maximum (as in the
diagram on the right). If that is the case then,
assuming that the input voltage is defined as
Vp Sin , the area under the curve between /2
and 3/2 radians will be the same as the area
under the curve between 3/2 and 5/2
radians (where 5/2 is the same as /2 in the
next cycle of the waveform). It follows that
the average current can be found by
Fig. 15.2
integrating the instantaneous current between
the limits /2 and 3/2 and dividing the result
by 3/2 - /2 = . Thus:
99
1
Iav =
3/2
Id d
(15.1)
/2
This change of limits reduces the computational burden of numerical integration by a factor of 2.
A further reduction in the computational overhead can be had by noting that, when a diode with
infinite parallel resistance is reverse biased, the current is limited to the reverse-saturation leakage
current. Moreover, it can be shown by simulation (e.g., by playing with the numbers in the
spreadsheet det_models.ods, sheet 2), that for a diode with Rds = 0 , the saturation current (-IS ) is
just reached when the diode reverse bias is the negative of the forward voltage drop under dynamic
conditions, i.e., -Vf . Since we have to input values of the peak and the DC output voltages (Vp and
Vm ) in order to perform the integration, we have a value of Vf = Vp - Vm that can be used to find
the part of the cycle in which the current is constant. Sensible values of Rds make very little
difference to the point at which -IS is reached; but even if the value chosen is unusually large, the
effect will be to increase Vf and thereby reduce the length of the region in which we assume that
Id = -IS .
To find the points at which the diode either comes out of or goes into the reverse saturation, we note
that this transition occurs when:
Vp Sinx - Vm = -Vf
Where x has two solutions because there are two crossover points in each cycle of the driving
waveform. Substituting for Vp :
( Vm + Vf ) Sinx - Vm = -Vf
( Vm + Vf ) Sinx = Vm - Vf
Sinx = ( Vm - Vf ) / ( Vm + Vf )
Thus in the region between 0 and /2 we have:
x = Arcsin[ ( Vm - Vf ) / ( Vm + Vf ) ]
Since we need to integrate between /2 and 3/2 however, the solution we want is:
x = - Arcsin[ ( Vm - Vf ) / ( Vm + Vf ) ]
15.2
Iav =
Id d +
Id d
/2
x
(15.3)
with the possibility of giving a fairly accurate solution for the second integral by simply setting its
100
indefinite value to -IS . Thus, applying the limits:
x
Iav =
Id d - (3/2 - x ) IS
/2
1
(15.4)
There is however, a good reason why using this form might not be a good idea. This is that the
solution for the diode forward voltage, Vf = Vp - Vm , has to be determined by a process of iteration.
Thus, when the integration routine is called at the beginning of the iteration process, the submitted
values of Vp and Vm might not be very accurate. If that is so, then the crossover point will not be
accurate, and the assumption that the function is flat from x to 3/2 might not be valid. Thus actual
evaluation of the second integral in (15.3) is advisable. There will still be a large saving in
computation time however because, as we will see, the numerical integration process can be made
adaptive, so that the integration of the nearly-flat region will be much faster than the integration of
the serpentine region from /2 to x .
(b - a)
f(x) dx
6
f(a) + 4f
( )
a+b
2
+ f(b)
(15.5)
The second-order approximation to the integral turns out to be the 1:4:1 weighted average of the
function evaluated at the lower limit, the mid point, and the upper limit, of the segment. If the
integrand is second-order and continuous across the interval, then the solution is exact. Otherwise
the value returned is an approximation, which improves as the difference between the limits a and b
is reduced.
It is possible to fit the curve segments to functions of even higher order. In fact, the trapezoid
83 http://en.wikipedia.org/wiki/Simpson's_rule
101
rule and Simpson's 1:4:1 rule are the first two in a series known as the 'Closed Newton-Cotes
formulae' 84. The cubic method is Simpson's 1:3:3:1 rule, and the quartic method is Boole's rule,
which uses the 7:32:12:32:7 weighted average; the function being evaluated at equally-spaced
intervals over the integration range in each case. Notice that the function always has to be
evaluated at p+1 points, where p is the order of the approximating function.
As the order of the fitting function is increased, the number of segments required to achieve a
given degree of accuracy is reduced. Increasing the order however, requires increasing the number
of points to be evaluated across a segment. Consequently there is not a linearly-increasing
advantage in increasing the order in comparison to the simple expedient of using more segments.
The greatest benefit is obtained by going from first-order to second-order, and for that reason,
Simpson's 1:4:1 rule is the most widely-used method. Another advantage of the 1:4:1 rule is that it
can be implemented in a manner that automatically determines the number of segments required
without the need to use arrays to store function-evaluation data.
Most code examples for Simpson's-rule integration require the number of segments to be given
as an input parameter. Presumably, the user will have performed an exhaustive evaluation of the
entire working parameter-space prior to using the method to solve actual problems, or will have
some other clairvoyant way of knowing that the result will be sufficiently accurate with a particular
choice. The 1:4:1 weighted average however, is convenient for an algorithm that successively
doubles the number of segments without either wasting the previous calculations or needing to store
them explicitly85. The new result after each doubling is checked against the previous result, and the
process continues until convergence occurs. In this way, Simpson's-rule integration becomes
adaptive, and advance characterisation of the integrand is not required.
The version of the adaptive algorithm used here works as follows. Firstly, a crude estimate of the
integral is obtained by applying Simpson's rule across the entire integration range. This is shown in
the adjacent diagram, where the interval between the three samples of f(x) is designated h0, so that:
h0 = (b-a)/2
If we call this initial estimate S0 , then we get
(by comparison with equation 15.5):
S0 = [ f(a) + 4f(a+h0 ) + f(a+2h0 ) ] h0 / 3
We can however, also de-construct the
problem in a different way by designating the samples as odd or even according to the number of
instances of the sampling interval that have been added to the lower limit. Then we can write the
sums of odds and evens separately. For the initial estimate we get:
0(even) = f(a) + f(a+2h0 )
0(odd) = f(a+h0)
so that:
S0 = [ 0(even) + 40(odd) ] h0 / 3
84 http://en.wikipedia.org/wiki/Newton-Cotes_formulas
85 The author was introduced to this method by Bob Weaver, who sent a code example in a private e-mail
communication, 26th April 2010, 06:20. Bob also describes the method and gives examples on his 'Numerical
Methods for Inductance Calculation' web page: http://electronbunker.ca/CalcMethods2c.html (accessed 25th Sept.
2014)
102
103
cycle no.
k
0
1
2
3
4
k
No. of
segs.
1
2
4
8
16
2k
odd evaluation
points, a + i hk
i=1
i=1,3
i=1,3,5,7
i = 1 , 3, 5 , . . . . , 15
i = 1 , 3, 5 , . . . . , 31
i = 1 , 3, 5 , . . . . , nk
n = max. value of i
1 = 2(0+1)-1
3 = 2(1+1)-1
7 = 2(3+1)-1
15 = 2(7+1)-1
31 = 2(15+1)-1
nk = 2( n(k-1)+1)-1
No, of segs
used
nsegs = ( nk+1) /2
The sum of odd samples can be calculated in a for-next loop, with a step size of 2. As can be seen
from the table, the maximum loop count for the kth iteration, nk , can be calculated from the
maximum loop count for the (k-1)th iteration using:
nk = 2 ( n(k-1) + 1 ) - 1
The termination criterion can be based on |Sn-Sn-1| falling below a certain value, or |1-Sn/Sn-1|
falling below a certain value, depending on the need for an absolute or a relative measure of
convergence.
The total number of segments used for a calculation can be determined from the final maximum
loop count using:
nsegs = ( nk+1) /2
The total number of points calculated is 2nsegs +1 = nk+2
A numerical solution for diode average current when Rds is finite, using the adaptive method just
described, is given below. For Id values, it makes successive calls to the diode instantaneous current
routine DinstI( ) described in section 15.1. The form of the integral is as per equation (15.1)
104
Function DavI(byval Vp as double, Vm as double, mVT as double, _
Isat as double, Rds as double) as double
'Calculates average diode current when diode has finite series resistance (Rds)
'Calls function DinstI( ). 'Adaptive Simpson's rule integration procedure
'based on a code example supplied by Bob Weaver (eml 2010:04:26 06:20).
Dim a as double, h as double, sumev as double, sumod as double
Dim previous as double, integral as double, diff as double
Dim n as integer, i as integer
a = pi/2
h = pi/2
sumev = DinstI(Vp*sin(a)-Vm, Rds, mVT, Isat) + DinstI(Vp*sin(3*a)-Vm, Rds, mVT, Isat)
sumod = DinstI(Vp*sin(2*a)-Vm, Rds, mVT, Isat)
previous = (4*sumod+sumev)*h/3
n=1
do
sumev = sumev + 2*sumod
sumod = 0
n = 2*(n+1)-1
h = h/2
for i = 1 to n step 2
sumod = sumod + DinstI(Vp*sin(a+i*h)-Vm, Rds, mVT, Isat)
next
integral = (4*sumod + sumev)*h/3
diff = integral / previous -1
previous = integral
loop until abs(diff) <= 1E-9
DavI = integral / pi
end function
Note that for the simple detector case, Iav can be calculated from ( Vp-Vf )/ RD , where Vf is obtained
using the routine DVfp2m( RD ; Vp ; mVT ; IS ) (see section 13.3). If Rds is set to zero in the call to
DavI( Vp ; Vm ; mVT ; IS ; Rds ) above, and the convergence criterion of 1 part in 109 change since
last cycle of iteration (as shown) is used, then, for practical input voltages, the numerical method
uses 64 segments and agrees with the Bessel function method to better than 9 decimal places of
micro amps (see spreadsheet det_models.ods, sheet 2).
105
15.3 Output voltage from peak input voltage
The problem of calculating the output voltage. Vm , from a given peak input voltage for the simple
detector was addressed in section 13.3. An iterative approach is required because calculation of the
diode average current, from which the output voltage is obtained, requires values of both the peak
input voltage and the output voltage. Thus it is necessary to input trial values of Vm and adjust
them until the calculation agrees with the input. The same conditions prevail when the diode has
non-zero series resistance, although the practicalities are somewhat different.
Using the diode average current routine developed in the previous subsections we have:
Vm = Iav RD = RD DavI( Vp ; Vm ; mVT ; IS ; Rds )
If we call the two instances of Vm y and x, then:
y = Iav RD = RD DavI( Vp ; x ; mVT ; IS ; Rds )
A good first estimate for x can be obtained by assuming that Rds = 0 and using the routine
DVfp2m( RD ; Vp ; mVT ; IS ). This works very well for small inputs, because the voltage across Rds
makes little contribution to the overall forward voltage when the diode current is small. Subsequent
adjustments to x however need to be determined with some precision, because a small change in x
gives rise to a large change in y, and the calculation can easily become unstable. We therefore need
to use the derivative y/x to determine the shift, with the small difficulty that this derivative cannot
be expressed analytically. A solution is to use finite-difference numerical differentiation.
y/x RD [ DavI( Vp ; x+x/2 ; mVT ; IS ; Rds ) + DavI( Vp ; x-x/2 ; mVT ; IS ; Rds ) ] / x
The finite difference, x , is determined by trial and error, and in this case, values between 10-3 and
10-6 were found to work well. Notice incidentally, that some implementations of finite-difference
differentiations use f(x) and f(x+x) respectively for the two function evaluations. This saves
computation time, because f(x) will already have been calculated, but it also reduces the accuracy of
the result and introduces bias into any residual error that remains when the calculation is deemed to
have converged.
Once we have a derivative, calculation of the shift is as in the various similar procedures in the set
of detector utilities accompanying this work:
x ( x - y ) / ( y/x - 1)
A Basic routine that performs the calculations given below. Note that it returns the diode forward
voltage drop Vf , which can be subtracted from the peak input voltage to determine the detector DC
output.
Regarding the matter of computational efficiency; notice that the function makes 3 calls to the
numerical integration routine DavI( ) on each round of iteration. DavI( ) in turn, if it uses say 64
segments, will make 127 calls on the instantaneous current routine DinstI( ) each time it is called.
When using interpreted Basic, execution is is slow; except for small input voltages, in which case
Rds makes very little difference to the result and the initial estimate is close to the final value.
106
>>>>>>>>>>>> work in progress
The procedure below has a numerical instability. The problem has yet to be solved.
see spreadsheet det_models.ods, sheet 9
>>>>>
107
15.4 Comparison of numerical integration and transformation methods
108
15.x Peak input voltage from output voltage
Works down to 1 microvolt with dx=0.00001
reducing dx beyond that causes no. of segs used in DavI() to exceed maxint (32767).
Note that there is little to be gained from including Rds in the model when Vm is very small.
Adaptive termination criterion is needed to avoid stability issues. Forced termination for n>256
was required during debugging to prevent infinite looping. Leaving it in removes the risk of
'program not responding' errors when termination criterion is adjusted.
109
110
111
>>> Text from old web article
Diode correction function
When using a diode detector to make accurate RF voltage measurements, mathematical correction
for diode non-linearity can be obtained by fitting the diode forward conduction characteristic to an
expression of the form:
Vf = V1 ln( If ) + V0 + If Rds
Where Rds is the diode series resistance, V0 + If Rds is the forward voltage when ln( If ) = 0 , and V1
is the gradient of the corresponding graph. We can see how this simple correction function comes
about by starting with the diode equation and including a term to allow for the fact that the diode
will have some ordinary series resistance ( Rds ). Hence:
Vf = If Rds + m VT ln[( If / IS ) +1 ]
Vf has to go to zero when If = 0. Hence the +1 term inside the logarithm bracket is there to set a
boundary condition, which is:
ln[(If / IS ) +1] 0 as If 0
IS for a Schottky signal diode is usually somewhere around 10-9 Amps (i.e., 1nA). Hence, when the
detector is used with inputs somewhat greater than the forward conduction threshold, then
If / IS >> 1, and we can neglect the +1 term without noticeable effect. This leads to the large input
approximation:
Vf = If Rds + mVT ln( If / IS )
But we don't know IS , and so we perform a substitution to capture this lack of knowledge in a
dimensionless parameter ( Iref / IS ). Thus:
Vf = If Rds + mVT ln[ ( If / Iref )( Iref / IS ) ]
Now, making use of the identity ln(pq)=ln(p)+ln(q), we get:
Vf = If Rds + mVT ln(Iref / IS) + mVT ln(If / Iref )
which is in the form:
Vf = If Rds + V0 + V1 ln( If / Iref )
The choice of Iref is arbitrary, and it is sensible to use some convenient engineering multiple of
Amps (i.e., usually 1 A or 1 mA for signal diodes). Hence, adopting Iref = 1 A :
Vf = If Rds + V0 + V1 ln( If / [A] )
i.e., the current inserted into the logarithm bracket has effectively been divided by its units to make
it dimensionless, which means that all we do in practice is take the logarithm of the forward current
in A to calculate the rightmost term. Now we can rearrange the equation in the form y=a+bx and
112
carry out a linear regression analysis86, i.e.;
Vf - If Rds = V0 + V1 ln( If )
where y = ( Vf - If Rds) , a = V0 , b = V1 and
x = ln( If )
When calibrating a small-signal detector for currents of < 1mA, Rds can usually be assumed to be
zero if a Schottky or gold-bonded diode is used. The series resistance of germanium point-contact
diodes is fairly large however and cannot be ignored. In order to allow for a finite Rds ; a simple
trick in calculation (program or spreadsheet) is to make provision for the term If Rds to be subtracted
from the measured value of Vf , but initially to set Rds = 0. If the regression-line shows significant
curvature, Rds can be adjusted by hand or by iteration to get the smallest standard-deviation of fit.
Once the parameters V0 , V1 and Rds are determined, the function we started with:
Vf = V1 ln( If ) + V0 + If Rds
will return a value of Vf for a given value of If that is good over several decades of current
(provided that the temperature is close to what it was when the fitting data were collected). Note
however, that the regression function does not have the correct boundary condition to return a true
value for Vf when If 0; and so If of less than about 100 nA should be trapped as an illegal input.
Alternatively we can solve for the diode equation parameters using:
V1 = mVT and V0 = mVT ln( Iref / IS )
in which case Vf can be calculated from the diode equation directly and the boundary condition at
If = 0 will be correct..
In an example given elsewhere [see Data Analysis], the regression function for a 1N5711 diode
(neglecting Rds ) was found to be:
Vf = 0.158342 + 0.029060 ln( If / [A]) Volts
Taking VT =25.3mV, this gives m=1.15 and IS = 4.3nA. Using the diode equation (with the
parameters in full precision as calculated from the fit) produces Vf values that are barely different
from those given by the fitting function provided that If >>IS (see spreadsheet 1N5711.ods). Note
that the determined diode equation parameters are not necessarily realistic, because Rds had been
neglected in this case, but they are accurate for several more decimal places than are required for
correcting experimental detector readings (The SPICE parameters for the Agilent 1N5711
[1n5711data.pdf] are IS = 2.2 nA. Rds =25 ). Attempting to include the If Rds correction in the fit
resulted in a negative value for Rds, indicating that there are insufficient data to determine the extra
parameter in this case.